bedrock: Model streamlining and cleanup (#49287)

Shardul Vaidya and Ona created 1 month ago

Release Notes:

- Improved Bedrock error messages: region-locked models ask the user to
try a different region, rate limits and access errors are reported
cleanly instead of as raw API responses
- Streamlined Bedrock model list to 39 curated models
- Fixed API errors when using non-tool models in agent threads

---------

Co-authored-by: Ona <[e-mail]>

Change summary

crates/bedrock/src/bedrock.rs                  |  50 
crates/bedrock/src/models.rs                   | 933 ++++++++-----------
crates/language_models/src/provider/bedrock.rs | 123 +
3 files changed, 526 insertions(+), 580 deletions(-)

Detailed changes

crates/bedrock/src/bedrock.rs

@@ -1,6 +1,6 @@
 mod models;
 
-use anyhow::{Context, Error, Result, anyhow};
+use anyhow::{Result, anyhow};
 use aws_sdk_bedrockruntime as bedrock;
 pub use aws_sdk_bedrockruntime as bedrock_client;
 use aws_sdk_bedrockruntime::types::InferenceConfiguration;
@@ -37,7 +37,7 @@ pub const CONTEXT_1M_BETA_HEADER: &str = "context-1m-2025-08-07";
 pub async fn stream_completion(
     client: bedrock::Client,
     request: Request,
-) -> Result<BoxStream<'static, Result<BedrockStreamingResponse, BedrockError>>, Error> {
+) -> Result<BoxStream<'static, Result<BedrockStreamingResponse, anyhow::Error>>, BedrockError> {
     let mut response = bedrock::Client::converse_stream(&client)
         .model_id(request.model.clone())
         .set_messages(request.messages.into());
@@ -94,10 +94,30 @@ pub async fn stream_completion(
         }
     }
 
-    let output = response
-        .send()
-        .await
-        .context("Failed to send API request to Bedrock");
+    let output = response.send().await.map_err(|err| match err {
+        bedrock::error::SdkError::ServiceError(ctx) => {
+            use bedrock::operation::converse_stream::ConverseStreamError;
+            let err = ctx.into_err();
+            match &err {
+                ConverseStreamError::ValidationException(e) => {
+                    BedrockError::Validation(e.message().unwrap_or("validation error").to_string())
+                }
+                ConverseStreamError::ThrottlingException(_) => BedrockError::RateLimited,
+                ConverseStreamError::ServiceUnavailableException(_)
+                | ConverseStreamError::ModelNotReadyException(_) => {
+                    BedrockError::ServiceUnavailable
+                }
+                ConverseStreamError::AccessDeniedException(e) => {
+                    BedrockError::AccessDenied(e.message().unwrap_or("access denied").to_string())
+                }
+                ConverseStreamError::InternalServerException(e) => BedrockError::InternalServer(
+                    e.message().unwrap_or("internal server error").to_string(),
+                ),
+                _ => BedrockError::Other(err.into()),
+            }
+        }
+        other => BedrockError::Other(other.into()),
+    });
 
     let stream = Box::pin(stream::unfold(
         output?.stream,
@@ -106,10 +126,10 @@ pub async fn stream_completion(
                 Ok(Some(output)) => Some((Ok(output), stream)),
                 Ok(None) => None,
                 Err(err) => Some((
-                    Err(BedrockError::ClientError(anyhow!(
+                    Err(anyhow!(
                         "{}",
                         aws_sdk_bedrockruntime::error::DisplayErrorContext(err)
-                    ))),
+                    )),
                     stream,
                 )),
             }
@@ -196,10 +216,16 @@ pub struct Metadata {
 
 #[derive(Error, Debug)]
 pub enum BedrockError {
-    #[error("client error: {0}")]
-    ClientError(anyhow::Error),
-    #[error("extension error: {0}")]
-    ExtensionError(anyhow::Error),
+    #[error("{0}")]
+    Validation(String),
+    #[error("rate limited")]
+    RateLimited,
+    #[error("service unavailable")]
+    ServiceUnavailable,
+    #[error("{0}")]
+    AccessDenied(String),
+    #[error("{0}")]
+    InternalServer(String),
     #[error(transparent)]
     Other(#[from] anyhow::Error),
 }

crates/bedrock/src/models.rs

@@ -45,7 +45,9 @@ pub struct BedrockModelCacheConfiguration {
 #[cfg_attr(feature = "schemars", derive(schemars::JsonSchema))]
 #[derive(Clone, Debug, Default, Serialize, Deserialize, PartialEq, EnumIter)]
 pub enum Model {
-    // Anthropic models (already included)
+    // Anthropic Claude 4+ models
+    #[serde(rename = "claude-haiku-4-5", alias = "claude-haiku-4-5-latest")]
+    ClaudeHaiku4_5,
     #[serde(rename = "claude-sonnet-4", alias = "claude-sonnet-4-latest")]
     ClaudeSonnet4,
     #[serde(
@@ -61,15 +63,8 @@ pub enum Model {
         alias = "claude-sonnet-4-5-thinking-latest"
     )]
     ClaudeSonnet4_5Thinking,
-    #[serde(rename = "claude-opus-4", alias = "claude-opus-4-latest")]
-    ClaudeOpus4,
     #[serde(rename = "claude-opus-4-1", alias = "claude-opus-4-1-latest")]
     ClaudeOpus4_1,
-    #[serde(
-        rename = "claude-opus-4-thinking",
-        alias = "claude-opus-4-thinking-latest"
-    )]
-    ClaudeOpus4Thinking,
     #[serde(
         rename = "claude-opus-4-1-thinking",
         alias = "claude-opus-4-1-thinking-latest"
@@ -89,79 +84,83 @@ pub enum Model {
         alias = "claude-opus-4-6-thinking-latest"
     )]
     ClaudeOpus4_6Thinking,
-    #[serde(rename = "claude-3-5-sonnet-v2", alias = "claude-3-5-sonnet-latest")]
-    Claude3_5SonnetV2,
-    #[serde(rename = "claude-3-7-sonnet", alias = "claude-3-7-sonnet-latest")]
-    Claude3_7Sonnet,
-    #[serde(
-        rename = "claude-3-7-sonnet-thinking",
-        alias = "claude-3-7-sonnet-thinking-latest"
-    )]
-    Claude3_7SonnetThinking,
-    #[serde(rename = "claude-3-opus", alias = "claude-3-opus-latest")]
-    Claude3Opus,
-    #[serde(rename = "claude-3-sonnet", alias = "claude-3-sonnet-latest")]
-    Claude3Sonnet,
-    #[serde(rename = "claude-3-5-haiku", alias = "claude-3-5-haiku-latest")]
-    Claude3_5Haiku,
-    #[serde(rename = "claude-haiku-4-5", alias = "claude-haiku-4-5-latest")]
-    ClaudeHaiku4_5,
-    Claude3_5Sonnet,
-    Claude3Haiku,
-    // Amazon Nova Models
-    AmazonNovaLite,
-    AmazonNovaMicro,
-    AmazonNovaPro,
-    AmazonNovaPremier,
-    // AI21 models
-    AI21J2GrandeInstruct,
-    AI21J2JumboInstruct,
-    AI21J2Mid,
-    AI21J2MidV1,
-    AI21J2Ultra,
-    AI21J2UltraV1_8k,
-    AI21J2UltraV1,
-    AI21JambaInstructV1,
-    AI21Jamba15LargeV1,
-    AI21Jamba15MiniV1,
-    // Cohere models
-    CohereCommandTextV14_4k,
-    CohereCommandRV1,
-    CohereCommandRPlusV1,
-    CohereCommandLightTextV14_4k,
-    // DeepSeek
-    DeepSeekR1,
-    // Meta models
-    MetaLlama38BInstructV1,
-    MetaLlama370BInstructV1,
-    MetaLlama318BInstructV1_128k,
-    MetaLlama318BInstructV1,
-    MetaLlama3170BInstructV1_128k,
-    MetaLlama3170BInstructV1,
-    MetaLlama31405BInstructV1,
-    MetaLlama321BInstructV1,
-    MetaLlama323BInstructV1,
-    MetaLlama3211BInstructV1,
-    MetaLlama3290BInstructV1,
-    MetaLlama3370BInstructV1,
-    #[allow(non_camel_case_types)]
-    MetaLlama4Scout17BInstructV1,
-    #[allow(non_camel_case_types)]
-    MetaLlama4Maverick17BInstructV1,
+
+    // Meta Llama 4 models
+    #[serde(rename = "llama-4-scout-17b")]
+    Llama4Scout17B,
+    #[serde(rename = "llama-4-maverick-17b")]
+    Llama4Maverick17B,
+
+    // Google Gemma 3 models
+    #[serde(rename = "gemma-3-4b")]
+    Gemma3_4B,
+    #[serde(rename = "gemma-3-12b")]
+    Gemma3_12B,
+    #[serde(rename = "gemma-3-27b")]
+    Gemma3_27B,
+
     // Mistral models
-    MistralMistral7BInstructV0,
-    MistralMixtral8x7BInstructV0,
-    MistralMistralLarge2402V1,
-    MistralMistralSmall2402V1,
-    MistralPixtralLarge2502V1,
-    // Writer models
-    PalmyraWriterX5,
-    PalmyraWriterX4,
+    #[serde(rename = "magistral-small")]
+    MagistralSmall,
+    #[serde(rename = "mistral-large-3")]
+    MistralLarge3,
+    #[serde(rename = "pixtral-large")]
+    PixtralLarge,
+
+    // Qwen models
+    #[serde(rename = "qwen3-32b")]
+    Qwen3_32B,
+    #[serde(rename = "qwen3-vl-235b")]
+    Qwen3VL235B,
+    #[serde(rename = "qwen3-235b")]
+    Qwen3_235B,
+    #[serde(rename = "qwen3-next-80b")]
+    Qwen3Next80B,
+    #[serde(rename = "qwen3-coder-30b")]
+    Qwen3Coder30B,
+    #[serde(rename = "qwen3-coder-next")]
+    Qwen3CoderNext,
+    #[serde(rename = "qwen3-coder-480b")]
+    Qwen3Coder480B,
+
+    // Amazon Nova models
+    #[serde(rename = "nova-lite")]
+    NovaLite,
+    #[serde(rename = "nova-pro")]
+    NovaPro,
+    #[serde(rename = "nova-premier")]
+    NovaPremier,
+    #[serde(rename = "nova-2-lite")]
+    Nova2Lite,
+
+    // OpenAI GPT OSS models
+    #[serde(rename = "gpt-oss-20b")]
+    GptOss20B,
+    #[serde(rename = "gpt-oss-120b")]
+    GptOss120B,
+
+    // MiniMax models
+    #[serde(rename = "minimax-m2")]
+    MiniMaxM2,
+
+    // Moonshot models
+    #[serde(rename = "kimi-k2-thinking")]
+    KimiK2Thinking,
+    #[serde(rename = "kimi-k2-5")]
+    KimiK2_5,
+
+    // DeepSeek models
+    #[serde(rename = "deepseek-r1")]
+    DeepSeekR1,
+    #[serde(rename = "deepseek-v3")]
+    DeepSeekV3_1,
+    #[serde(rename = "deepseek-v3-2")]
+    DeepSeekV3_2,
+
     #[serde(rename = "custom")]
     Custom {
         name: String,
         max_tokens: u64,
-        /// The name displayed in the UI, such as in the assistant panel model dropdown menu.
         display_name: Option<String>,
         max_output_tokens: Option<u64>,
         default_temperature: Option<f32>,
@@ -170,12 +169,8 @@ pub enum Model {
 }
 
 impl Model {
-    pub fn default_fast(region: &str) -> Self {
-        if region.starts_with("us-") {
-            Self::Claude3_5Haiku
-        } else {
-            Self::Claude3Haiku
-        }
+    pub fn default_fast(_region: &str) -> Self {
+        Self::ClaudeHaiku4_5
     }
 
     pub fn from_id(id: &str) -> anyhow::Result<Self> {
@@ -191,24 +186,6 @@ impl Model {
             Ok(Self::ClaudeOpus4_1Thinking)
         } else if id.starts_with("claude-opus-4-1") {
             Ok(Self::ClaudeOpus4_1)
-        } else if id.starts_with("claude-opus-4-thinking") {
-            Ok(Self::ClaudeOpus4Thinking)
-        } else if id.starts_with("claude-opus-4") {
-            Ok(Self::ClaudeOpus4)
-        } else if id.starts_with("claude-3-5-sonnet-v2") {
-            Ok(Self::Claude3_5SonnetV2)
-        } else if id.starts_with("claude-3-opus") {
-            Ok(Self::Claude3Opus)
-        } else if id.starts_with("claude-3-sonnet") {
-            Ok(Self::Claude3Sonnet)
-        } else if id.starts_with("claude-3-5-haiku") {
-            Ok(Self::Claude3_5Haiku)
-        } else if id.starts_with("claude-haiku-4-5") {
-            Ok(Self::ClaudeHaiku4_5)
-        } else if id.starts_with("claude-3-7-sonnet") {
-            Ok(Self::Claude3_7Sonnet)
-        } else if id.starts_with("claude-3-7-sonnet-thinking") {
-            Ok(Self::Claude3_7SonnetThinking)
         } else if id.starts_with("claude-sonnet-4-5-thinking") {
             Ok(Self::ClaudeSonnet4_5Thinking)
         } else if id.starts_with("claude-sonnet-4-5") {
@@ -217,6 +194,8 @@ impl Model {
             Ok(Self::ClaudeSonnet4Thinking)
         } else if id.starts_with("claude-sonnet-4") {
             Ok(Self::ClaudeSonnet4)
+        } else if id.starts_with("claude-haiku-4-5") {
+            Ok(Self::ClaudeHaiku4_5)
         } else {
             anyhow::bail!("invalid model id {id}");
         }
@@ -224,283 +203,228 @@ impl Model {
 
     pub fn id(&self) -> &str {
         match self {
-            Model::ClaudeSonnet4 => "claude-sonnet-4",
-            Model::ClaudeSonnet4Thinking => "claude-sonnet-4-thinking",
-            Model::ClaudeSonnet4_5 => "claude-sonnet-4-5",
-            Model::ClaudeSonnet4_5Thinking => "claude-sonnet-4-5-thinking",
-            Model::ClaudeOpus4 => "claude-opus-4",
-            Model::ClaudeOpus4_1 => "claude-opus-4-1",
-            Model::ClaudeOpus4Thinking => "claude-opus-4-thinking",
-            Model::ClaudeOpus4_1Thinking => "claude-opus-4-1-thinking",
-            Model::ClaudeOpus4_5 => "claude-opus-4-5",
-            Model::ClaudeOpus4_5Thinking => "claude-opus-4-5-thinking",
-            Model::ClaudeOpus4_6 => "claude-opus-4-6",
-            Model::ClaudeOpus4_6Thinking => "claude-opus-4-6-thinking",
-            Model::Claude3_5SonnetV2 => "claude-3-5-sonnet-v2",
-            Model::Claude3_5Sonnet => "claude-3-5-sonnet",
-            Model::Claude3Opus => "claude-3-opus",
-            Model::Claude3Sonnet => "claude-3-sonnet",
-            Model::Claude3Haiku => "claude-3-haiku",
-            Model::Claude3_5Haiku => "claude-3-5-haiku",
-            Model::ClaudeHaiku4_5 => "claude-haiku-4-5",
-            Model::Claude3_7Sonnet => "claude-3-7-sonnet",
-            Model::Claude3_7SonnetThinking => "claude-3-7-sonnet-thinking",
-            Model::AmazonNovaLite => "amazon-nova-lite",
-            Model::AmazonNovaMicro => "amazon-nova-micro",
-            Model::AmazonNovaPro => "amazon-nova-pro",
-            Model::AmazonNovaPremier => "amazon-nova-premier",
-            Model::DeepSeekR1 => "deepseek-r1",
-            Model::AI21J2GrandeInstruct => "ai21-j2-grande-instruct",
-            Model::AI21J2JumboInstruct => "ai21-j2-jumbo-instruct",
-            Model::AI21J2Mid => "ai21-j2-mid",
-            Model::AI21J2MidV1 => "ai21-j2-mid-v1",
-            Model::AI21J2Ultra => "ai21-j2-ultra",
-            Model::AI21J2UltraV1_8k => "ai21-j2-ultra-v1-8k",
-            Model::AI21J2UltraV1 => "ai21-j2-ultra-v1",
-            Model::AI21JambaInstructV1 => "ai21-jamba-instruct-v1",
-            Model::AI21Jamba15LargeV1 => "ai21-jamba-1-5-large-v1",
-            Model::AI21Jamba15MiniV1 => "ai21-jamba-1-5-mini-v1",
-            Model::CohereCommandTextV14_4k => "cohere-command-text-v14-4k",
-            Model::CohereCommandRV1 => "cohere-command-r-v1",
-            Model::CohereCommandRPlusV1 => "cohere-command-r-plus-v1",
-            Model::CohereCommandLightTextV14_4k => "cohere-command-light-text-v14-4k",
-            Model::MetaLlama38BInstructV1 => "meta-llama3-8b-instruct-v1",
-            Model::MetaLlama370BInstructV1 => "meta-llama3-70b-instruct-v1",
-            Model::MetaLlama318BInstructV1_128k => "meta-llama3-1-8b-instruct-v1-128k",
-            Model::MetaLlama318BInstructV1 => "meta-llama3-1-8b-instruct-v1",
-            Model::MetaLlama3170BInstructV1_128k => "meta-llama3-1-70b-instruct-v1-128k",
-            Model::MetaLlama3170BInstructV1 => "meta-llama3-1-70b-instruct-v1",
-            Model::MetaLlama31405BInstructV1 => "meta-llama3-1-405b-instruct-v1",
-            Model::MetaLlama321BInstructV1 => "meta-llama3-2-1b-instruct-v1",
-            Model::MetaLlama323BInstructV1 => "meta-llama3-2-3b-instruct-v1",
-            Model::MetaLlama3211BInstructV1 => "meta-llama3-2-11b-instruct-v1",
-            Model::MetaLlama3290BInstructV1 => "meta-llama3-2-90b-instruct-v1",
-            Model::MetaLlama3370BInstructV1 => "meta-llama3-3-70b-instruct-v1",
-            Model::MetaLlama4Scout17BInstructV1 => "meta-llama4-scout-17b-instruct-v1",
-            Model::MetaLlama4Maverick17BInstructV1 => "meta-llama4-maverick-17b-instruct-v1",
-            Model::MistralMistral7BInstructV0 => "mistral-7b-instruct-v0",
-            Model::MistralMixtral8x7BInstructV0 => "mistral-mixtral-8x7b-instruct-v0",
-            Model::MistralMistralLarge2402V1 => "mistral-large-2402-v1",
-            Model::MistralMistralSmall2402V1 => "mistral-small-2402-v1",
-            Model::MistralPixtralLarge2502V1 => "mistral-pixtral-large-2502-v1",
-            Model::PalmyraWriterX4 => "palmyra-writer-x4",
-            Model::PalmyraWriterX5 => "palmyra-writer-x5",
+            Self::ClaudeHaiku4_5 => "claude-haiku-4-5",
+            Self::ClaudeSonnet4 => "claude-sonnet-4",
+            Self::ClaudeSonnet4Thinking => "claude-sonnet-4-thinking",
+            Self::ClaudeSonnet4_5 => "claude-sonnet-4-5",
+            Self::ClaudeSonnet4_5Thinking => "claude-sonnet-4-5-thinking",
+            Self::ClaudeOpus4_1 => "claude-opus-4-1",
+            Self::ClaudeOpus4_1Thinking => "claude-opus-4-1-thinking",
+            Self::ClaudeOpus4_5 => "claude-opus-4-5",
+            Self::ClaudeOpus4_5Thinking => "claude-opus-4-5-thinking",
+            Self::ClaudeOpus4_6 => "claude-opus-4-6",
+            Self::ClaudeOpus4_6Thinking => "claude-opus-4-6-thinking",
+            Self::Llama4Scout17B => "llama-4-scout-17b",
+            Self::Llama4Maverick17B => "llama-4-maverick-17b",
+            Self::Gemma3_4B => "gemma-3-4b",
+            Self::Gemma3_12B => "gemma-3-12b",
+            Self::Gemma3_27B => "gemma-3-27b",
+            Self::MagistralSmall => "magistral-small",
+            Self::MistralLarge3 => "mistral-large-3",
+            Self::PixtralLarge => "pixtral-large",
+            Self::Qwen3_32B => "qwen3-32b",
+            Self::Qwen3VL235B => "qwen3-vl-235b",
+            Self::Qwen3_235B => "qwen3-235b",
+            Self::Qwen3Next80B => "qwen3-next-80b",
+            Self::Qwen3Coder30B => "qwen3-coder-30b",
+            Self::Qwen3CoderNext => "qwen3-coder-next",
+            Self::Qwen3Coder480B => "qwen3-coder-480b",
+            Self::NovaLite => "nova-lite",
+            Self::NovaPro => "nova-pro",
+            Self::NovaPremier => "nova-premier",
+            Self::Nova2Lite => "nova-2-lite",
+            Self::GptOss20B => "gpt-oss-20b",
+            Self::GptOss120B => "gpt-oss-120b",
+            Self::MiniMaxM2 => "minimax-m2",
+            Self::KimiK2Thinking => "kimi-k2-thinking",
+            Self::KimiK2_5 => "kimi-k2-5",
+            Self::DeepSeekR1 => "deepseek-r1",
+            Self::DeepSeekV3_1 => "deepseek-v3",
+            Self::DeepSeekV3_2 => "deepseek-v3-2",
             Self::Custom { name, .. } => name,
         }
     }
 
     pub fn request_id(&self) -> &str {
         match self {
-            Model::ClaudeSonnet4 | Model::ClaudeSonnet4Thinking => {
+            Self::ClaudeHaiku4_5 => "anthropic.claude-haiku-4-5-20251001-v1:0",
+            Self::ClaudeSonnet4 | Self::ClaudeSonnet4Thinking => {
                 "anthropic.claude-sonnet-4-20250514-v1:0"
             }
-            Model::ClaudeSonnet4_5 | Model::ClaudeSonnet4_5Thinking => {
+            Self::ClaudeSonnet4_5 | Self::ClaudeSonnet4_5Thinking => {
                 "anthropic.claude-sonnet-4-5-20250929-v1:0"
             }
-            Model::ClaudeOpus4 | Model::ClaudeOpus4Thinking => {
-                "anthropic.claude-opus-4-20250514-v1:0"
-            }
-            Model::ClaudeOpus4_1 | Model::ClaudeOpus4_1Thinking => {
+            Self::ClaudeOpus4_1 | Self::ClaudeOpus4_1Thinking => {
                 "anthropic.claude-opus-4-1-20250805-v1:0"
             }
-            Model::ClaudeOpus4_5 | Model::ClaudeOpus4_5Thinking => {
+            Self::ClaudeOpus4_5 | Self::ClaudeOpus4_5Thinking => {
                 "anthropic.claude-opus-4-5-20251101-v1:0"
             }
-            Model::ClaudeOpus4_6 | Model::ClaudeOpus4_6Thinking => "anthropic.claude-opus-4-6-v1",
-            Model::Claude3_5SonnetV2 => "anthropic.claude-3-5-sonnet-20241022-v2:0",
-            Model::Claude3_5Sonnet => "anthropic.claude-3-5-sonnet-20240620-v1:0",
-            Model::Claude3Opus => "anthropic.claude-3-opus-20240229-v1:0",
-            Model::Claude3Sonnet => "anthropic.claude-3-sonnet-20240229-v1:0",
-            Model::Claude3Haiku => "anthropic.claude-3-haiku-20240307-v1:0",
-            Model::Claude3_5Haiku => "anthropic.claude-3-5-haiku-20241022-v1:0",
-            Model::ClaudeHaiku4_5 => "anthropic.claude-haiku-4-5-20251001-v1:0",
-            Model::Claude3_7Sonnet | Model::Claude3_7SonnetThinking => {
-                "anthropic.claude-3-7-sonnet-20250219-v1:0"
-            }
-            Model::AmazonNovaLite => "amazon.nova-lite-v1:0",
-            Model::AmazonNovaMicro => "amazon.nova-micro-v1:0",
-            Model::AmazonNovaPro => "amazon.nova-pro-v1:0",
-            Model::AmazonNovaPremier => "amazon.nova-premier-v1:0",
-            Model::DeepSeekR1 => "deepseek.r1-v1:0",
-            Model::AI21J2GrandeInstruct => "ai21.j2-grande-instruct",
-            Model::AI21J2JumboInstruct => "ai21.j2-jumbo-instruct",
-            Model::AI21J2Mid => "ai21.j2-mid",
-            Model::AI21J2MidV1 => "ai21.j2-mid-v1",
-            Model::AI21J2Ultra => "ai21.j2-ultra",
-            Model::AI21J2UltraV1_8k => "ai21.j2-ultra-v1:0:8k",
-            Model::AI21J2UltraV1 => "ai21.j2-ultra-v1",
-            Model::AI21JambaInstructV1 => "ai21.jamba-instruct-v1:0",
-            Model::AI21Jamba15LargeV1 => "ai21.jamba-1-5-large-v1:0",
-            Model::AI21Jamba15MiniV1 => "ai21.jamba-1-5-mini-v1:0",
-            Model::CohereCommandTextV14_4k => "cohere.command-text-v14:7:4k",
-            Model::CohereCommandRV1 => "cohere.command-r-v1:0",
-            Model::CohereCommandRPlusV1 => "cohere.command-r-plus-v1:0",
-            Model::CohereCommandLightTextV14_4k => "cohere.command-light-text-v14:7:4k",
-            Model::MetaLlama38BInstructV1 => "meta.llama3-8b-instruct-v1:0",
-            Model::MetaLlama370BInstructV1 => "meta.llama3-70b-instruct-v1:0",
-            Model::MetaLlama318BInstructV1_128k => "meta.llama3-1-8b-instruct-v1:0",
-            Model::MetaLlama318BInstructV1 => "meta.llama3-1-8b-instruct-v1:0",
-            Model::MetaLlama3170BInstructV1_128k => "meta.llama3-1-70b-instruct-v1:0",
-            Model::MetaLlama3170BInstructV1 => "meta.llama3-1-70b-instruct-v1:0",
-            Model::MetaLlama31405BInstructV1 => "meta.llama3-1-405b-instruct-v1:0",
-            Model::MetaLlama3211BInstructV1 => "meta.llama3-2-11b-instruct-v1:0",
-            Model::MetaLlama3290BInstructV1 => "meta.llama3-2-90b-instruct-v1:0",
-            Model::MetaLlama321BInstructV1 => "meta.llama3-2-1b-instruct-v1:0",
-            Model::MetaLlama323BInstructV1 => "meta.llama3-2-3b-instruct-v1:0",
-            Model::MetaLlama3370BInstructV1 => "meta.llama3-3-70b-instruct-v1:0",
-            Model::MetaLlama4Scout17BInstructV1 => "meta.llama4-scout-17b-instruct-v1:0",
-            Model::MetaLlama4Maverick17BInstructV1 => "meta.llama4-maverick-17b-instruct-v1:0",
-            Model::MistralMistral7BInstructV0 => "mistral.mistral-7b-instruct-v0:2",
-            Model::MistralMixtral8x7BInstructV0 => "mistral.mixtral-8x7b-instruct-v0:1",
-            Model::MistralMistralLarge2402V1 => "mistral.mistral-large-2402-v1:0",
-            Model::MistralMistralSmall2402V1 => "mistral.mistral-small-2402-v1:0",
-            Model::MistralPixtralLarge2502V1 => "mistral.pixtral-large-2502-v1:0",
-            Model::PalmyraWriterX4 => "writer.palmyra-x4-v1:0",
-            Model::PalmyraWriterX5 => "writer.palmyra-x5-v1:0",
+            Self::ClaudeOpus4_6 | Self::ClaudeOpus4_6Thinking => "anthropic.claude-opus-4-6-v1",
+            Self::Llama4Scout17B => "meta.llama4-scout-17b-instruct-v1:0",
+            Self::Llama4Maverick17B => "meta.llama4-maverick-17b-instruct-v1:0",
+            Self::Gemma3_4B => "google.gemma-3-4b-it",
+            Self::Gemma3_12B => "google.gemma-3-12b-it",
+            Self::Gemma3_27B => "google.gemma-3-27b-it",
+            Self::MagistralSmall => "mistral.magistral-small-2509",
+            Self::MistralLarge3 => "mistral.mistral-large-3-675b-instruct",
+            Self::PixtralLarge => "mistral.pixtral-large-2502-v1:0",
+            Self::Qwen3VL235B => "qwen.qwen3-vl-235b-a22b",
+            Self::Qwen3_32B => "qwen.qwen3-32b-v1:0",
+            Self::Qwen3_235B => "qwen.qwen3-235b-a22b-2507-v1:0",
+            Self::Qwen3Next80B => "qwen.qwen3-next-80b-a3b",
+            Self::Qwen3Coder30B => "qwen.qwen3-coder-30b-a3b-v1:0",
+            Self::Qwen3CoderNext => "qwen.qwen3-coder-next",
+            Self::Qwen3Coder480B => "qwen.qwen3-coder-480b-a35b-v1:0",
+            Self::NovaLite => "amazon.nova-lite-v1:0",
+            Self::NovaPro => "amazon.nova-pro-v1:0",
+            Self::NovaPremier => "amazon.nova-premier-v1:0",
+            Self::Nova2Lite => "amazon.nova-2-lite-v1:0",
+            Self::GptOss20B => "openai.gpt-oss-20b-1:0",
+            Self::GptOss120B => "openai.gpt-oss-120b-1:0",
+            Self::MiniMaxM2 => "minimax.minimax-m2",
+            Self::KimiK2Thinking => "moonshot.kimi-k2-thinking",
+            Self::KimiK2_5 => "moonshotai.kimi-k2.5",
+            Self::DeepSeekR1 => "deepseek.r1-v1:0",
+            Self::DeepSeekV3_1 => "deepseek.v3-v1:0",
+            Self::DeepSeekV3_2 => "deepseek.v3.2",
             Self::Custom { name, .. } => name,
         }
     }
 
     pub fn display_name(&self) -> &str {
         match self {
+            Self::ClaudeHaiku4_5 => "Claude Haiku 4.5",
             Self::ClaudeSonnet4 => "Claude Sonnet 4",
             Self::ClaudeSonnet4Thinking => "Claude Sonnet 4 Thinking",
             Self::ClaudeSonnet4_5 => "Claude Sonnet 4.5",
             Self::ClaudeSonnet4_5Thinking => "Claude Sonnet 4.5 Thinking",
-            Self::ClaudeOpus4 => "Claude Opus 4",
             Self::ClaudeOpus4_1 => "Claude Opus 4.1",
-            Self::ClaudeOpus4Thinking => "Claude Opus 4 Thinking",
             Self::ClaudeOpus4_1Thinking => "Claude Opus 4.1 Thinking",
             Self::ClaudeOpus4_5 => "Claude Opus 4.5",
             Self::ClaudeOpus4_5Thinking => "Claude Opus 4.5 Thinking",
             Self::ClaudeOpus4_6 => "Claude Opus 4.6",
             Self::ClaudeOpus4_6Thinking => "Claude Opus 4.6 Thinking",
-            Self::Claude3_5SonnetV2 => "Claude 3.5 Sonnet v2",
-            Self::Claude3_5Sonnet => "Claude 3.5 Sonnet",
-            Self::Claude3Opus => "Claude 3 Opus",
-            Self::Claude3Sonnet => "Claude 3 Sonnet",
-            Self::Claude3Haiku => "Claude 3 Haiku",
-            Self::Claude3_5Haiku => "Claude 3.5 Haiku",
-            Self::ClaudeHaiku4_5 => "Claude Haiku 4.5",
-            Self::Claude3_7Sonnet => "Claude 3.7 Sonnet",
-            Self::Claude3_7SonnetThinking => "Claude 3.7 Sonnet Thinking",
-            Self::AmazonNovaLite => "Amazon Nova Lite",
-            Self::AmazonNovaMicro => "Amazon Nova Micro",
-            Self::AmazonNovaPro => "Amazon Nova Pro",
-            Self::AmazonNovaPremier => "Amazon Nova Premier",
+            Self::Llama4Scout17B => "Llama 4 Scout 17B",
+            Self::Llama4Maverick17B => "Llama 4 Maverick 17B",
+            Self::Gemma3_4B => "Gemma 3 4B",
+            Self::Gemma3_12B => "Gemma 3 12B",
+            Self::Gemma3_27B => "Gemma 3 27B",
+            Self::MagistralSmall => "Magistral Small",
+            Self::MistralLarge3 => "Mistral Large 3",
+            Self::PixtralLarge => "Pixtral Large",
+            Self::Qwen3VL235B => "Qwen3 VL 235B",
+            Self::Qwen3_32B => "Qwen3 32B",
+            Self::Qwen3_235B => "Qwen3 235B",
+            Self::Qwen3Next80B => "Qwen3 Next 80B",
+            Self::Qwen3Coder30B => "Qwen3 Coder 30B",
+            Self::Qwen3CoderNext => "Qwen3 Coder Next",
+            Self::Qwen3Coder480B => "Qwen3 Coder 480B",
+            Self::NovaLite => "Amazon Nova Lite",
+            Self::NovaPro => "Amazon Nova Pro",
+            Self::NovaPremier => "Amazon Nova Premier",
+            Self::Nova2Lite => "Amazon Nova 2 Lite",
+            Self::GptOss20B => "GPT OSS 20B",
+            Self::GptOss120B => "GPT OSS 120B",
+            Self::MiniMaxM2 => "MiniMax M2",
+            Self::KimiK2Thinking => "Kimi K2 Thinking",
+            Self::KimiK2_5 => "Kimi K2.5",
             Self::DeepSeekR1 => "DeepSeek R1",
-            Self::AI21J2GrandeInstruct => "AI21 Jurassic2 Grande Instruct",
-            Self::AI21J2JumboInstruct => "AI21 Jurassic2 Jumbo Instruct",
-            Self::AI21J2Mid => "AI21 Jurassic2 Mid",
-            Self::AI21J2MidV1 => "AI21 Jurassic2 Mid V1",
-            Self::AI21J2Ultra => "AI21 Jurassic2 Ultra",
-            Self::AI21J2UltraV1_8k => "AI21 Jurassic2 Ultra V1 8K",
-            Self::AI21J2UltraV1 => "AI21 Jurassic2 Ultra V1",
-            Self::AI21JambaInstructV1 => "AI21 Jamba Instruct",
-            Self::AI21Jamba15LargeV1 => "AI21 Jamba 1.5 Large",
-            Self::AI21Jamba15MiniV1 => "AI21 Jamba 1.5 Mini",
-            Self::CohereCommandTextV14_4k => "Cohere Command Text V14 4K",
-            Self::CohereCommandRV1 => "Cohere Command R V1",
-            Self::CohereCommandRPlusV1 => "Cohere Command R Plus V1",
-            Self::CohereCommandLightTextV14_4k => "Cohere Command Light Text V14 4K",
-            Self::MetaLlama38BInstructV1 => "Meta Llama 3 8B Instruct",
-            Self::MetaLlama370BInstructV1 => "Meta Llama 3 70B Instruct",
-            Self::MetaLlama318BInstructV1_128k => "Meta Llama 3.1 8B Instruct 128K",
-            Self::MetaLlama318BInstructV1 => "Meta Llama 3.1 8B Instruct",
-            Self::MetaLlama3170BInstructV1_128k => "Meta Llama 3.1 70B Instruct 128K",
-            Self::MetaLlama3170BInstructV1 => "Meta Llama 3.1 70B Instruct",
-            Self::MetaLlama31405BInstructV1 => "Meta Llama 3.1 405B Instruct",
-            Self::MetaLlama3211BInstructV1 => "Meta Llama 3.2 11B Instruct",
-            Self::MetaLlama3290BInstructV1 => "Meta Llama 3.2 90B Instruct",
-            Self::MetaLlama321BInstructV1 => "Meta Llama 3.2 1B Instruct",
-            Self::MetaLlama323BInstructV1 => "Meta Llama 3.2 3B Instruct",
-            Self::MetaLlama3370BInstructV1 => "Meta Llama 3.3 70B Instruct",
-            Self::MetaLlama4Scout17BInstructV1 => "Meta Llama 4 Scout 17B Instruct",
-            Self::MetaLlama4Maverick17BInstructV1 => "Meta Llama 4 Maverick 17B Instruct",
-            Self::MistralMistral7BInstructV0 => "Mistral 7B Instruct V0",
-            Self::MistralMixtral8x7BInstructV0 => "Mistral Mixtral 8x7B Instruct V0",
-            Self::MistralMistralLarge2402V1 => "Mistral Large 2402 V1",
-            Self::MistralMistralSmall2402V1 => "Mistral Small 2402 V1",
-            Self::MistralPixtralLarge2502V1 => "Pixtral Large 25.02 V1",
-            Self::PalmyraWriterX5 => "Writer Palmyra X5",
-            Self::PalmyraWriterX4 => "Writer Palmyra X4",
+            Self::DeepSeekV3_1 => "DeepSeek V3.1",
+            Self::DeepSeekV3_2 => "DeepSeek V3.2",
             Self::Custom {
                 display_name, name, ..
-            } => display_name.as_deref().unwrap_or(name),
+            } => display_name.as_deref().unwrap_or(name.as_str()),
         }
     }
 
     pub fn max_token_count(&self) -> u64 {
+        self.max_tokens()
+    }
+
+    pub fn max_tokens(&self) -> u64 {
         match self {
-            Self::Claude3_5SonnetV2
-            | Self::Claude3Opus
-            | Self::Claude3Sonnet
-            | Self::Claude3_5Haiku
-            | Self::ClaudeHaiku4_5
-            | Self::Claude3_7Sonnet
+            Self::ClaudeHaiku4_5
             | Self::ClaudeSonnet4
-            | Self::ClaudeOpus4
-            | Self::ClaudeOpus4_1
             | Self::ClaudeSonnet4Thinking
             | Self::ClaudeSonnet4_5
             | Self::ClaudeSonnet4_5Thinking
-            | Self::ClaudeOpus4Thinking
+            | Self::ClaudeOpus4_1
             | Self::ClaudeOpus4_1Thinking
             | Self::ClaudeOpus4_5
             | Self::ClaudeOpus4_5Thinking
             | Self::ClaudeOpus4_6
             | Self::ClaudeOpus4_6Thinking => 200_000,
-            Self::AmazonNovaPremier => 1_000_000,
-            Self::PalmyraWriterX5 => 1_000_000,
-            Self::PalmyraWriterX4 => 128_000,
+            Self::Llama4Scout17B | Self::Llama4Maverick17B => 128_000,
+            Self::Gemma3_4B | Self::Gemma3_12B | Self::Gemma3_27B => 128_000,
+            Self::MagistralSmall | Self::MistralLarge3 | Self::PixtralLarge => 128_000,
+            Self::Qwen3_32B
+            | Self::Qwen3VL235B
+            | Self::Qwen3_235B
+            | Self::Qwen3Next80B
+            | Self::Qwen3Coder30B
+            | Self::Qwen3CoderNext
+            | Self::Qwen3Coder480B => 128_000,
+            Self::NovaLite | Self::NovaPro => 300_000,
+            Self::NovaPremier => 1_000_000,
+            Self::Nova2Lite => 300_000,
+            Self::GptOss20B | Self::GptOss120B => 128_000,
+            Self::MiniMaxM2 => 128_000,
+            Self::KimiK2Thinking | Self::KimiK2_5 => 128_000,
+            Self::DeepSeekR1 | Self::DeepSeekV3_1 | Self::DeepSeekV3_2 => 128_000,
             Self::Custom { max_tokens, .. } => *max_tokens,
-            _ => 128_000,
         }
     }
 
     pub fn max_output_tokens(&self) -> u64 {
         match self {
-            Self::Claude3Opus | Self::Claude3Sonnet | Self::Claude3_5Haiku => 4_096,
-            Self::Claude3_7Sonnet | Self::Claude3_7SonnetThinking => 128_000,
-            Self::ClaudeSonnet4 | Self::ClaudeSonnet4Thinking => 64_000,
-            Self::ClaudeSonnet4_5
+            Self::ClaudeHaiku4_5
+            | Self::ClaudeSonnet4_5
             | Self::ClaudeSonnet4_5Thinking
-            | Self::ClaudeHaiku4_5
             | Self::ClaudeOpus4_5
             | Self::ClaudeOpus4_5Thinking => 64_000,
+            Self::ClaudeSonnet4 | Self::ClaudeSonnet4Thinking => 64_000,
+            Self::ClaudeOpus4_1 | Self::ClaudeOpus4_1Thinking => 32_000,
             Self::ClaudeOpus4_6 | Self::ClaudeOpus4_6Thinking => 128_000,
-            Self::ClaudeOpus4
-            | Self::ClaudeOpus4Thinking
-            | Self::ClaudeOpus4_1
-            | Self::ClaudeOpus4_1Thinking => 32_000,
-            Self::Claude3_5SonnetV2 | Self::PalmyraWriterX4 | Self::PalmyraWriterX5 => 8_192,
+            Self::Llama4Scout17B
+            | Self::Llama4Maverick17B
+            | Self::Gemma3_4B
+            | Self::Gemma3_12B
+            | Self::Gemma3_27B
+            | Self::MagistralSmall
+            | Self::MistralLarge3
+            | Self::PixtralLarge => 8_192,
+            Self::Qwen3_32B
+            | Self::Qwen3VL235B
+            | Self::Qwen3_235B
+            | Self::Qwen3Next80B
+            | Self::Qwen3Coder30B
+            | Self::Qwen3CoderNext
+            | Self::Qwen3Coder480B => 8_192,
+            Self::NovaLite | Self::NovaPro | Self::NovaPremier | Self::Nova2Lite => 5_000,
+            Self::GptOss20B | Self::GptOss120B => 16_000,
+            Self::MiniMaxM2 => 16_000,
+            Self::KimiK2Thinking | Self::KimiK2_5 => 16_000,
+            Self::DeepSeekR1 | Self::DeepSeekV3_1 | Self::DeepSeekV3_2 => 16_000,
             Self::Custom {
                 max_output_tokens, ..
             } => max_output_tokens.unwrap_or(4_096),
-            _ => 4_096,
         }
     }
 
     pub fn default_temperature(&self) -> f32 {
         match self {
-            Self::Claude3_5SonnetV2
-            | Self::Claude3Opus
-            | Self::Claude3Sonnet
-            | Self::Claude3_5Haiku
-            | Self::ClaudeHaiku4_5
-            | Self::Claude3_7Sonnet
-            | Self::ClaudeOpus4
-            | Self::ClaudeOpus4Thinking
+            Self::ClaudeHaiku4_5
+            | Self::ClaudeSonnet4
+            | Self::ClaudeSonnet4Thinking
+            | Self::ClaudeSonnet4_5
+            | Self::ClaudeSonnet4_5Thinking
             | Self::ClaudeOpus4_1
             | Self::ClaudeOpus4_1Thinking
             | Self::ClaudeOpus4_5
             | Self::ClaudeOpus4_5Thinking
             | Self::ClaudeOpus4_6
-            | Self::ClaudeOpus4_6Thinking
-            | Self::ClaudeSonnet4
-            | Self::ClaudeSonnet4Thinking
-            | Self::ClaudeSonnet4_5
-            | Self::ClaudeSonnet4_5Thinking => 1.0,
+            | Self::ClaudeOpus4_6Thinking => 1.0,
             Self::Custom {
                 default_temperature,
                 ..
@@ -511,127 +435,97 @@ impl Model {
 
     pub fn supports_tool_use(&self) -> bool {
         match self {
-            // Anthropic Claude 3 models (all support tool use)
-            Self::Claude3Opus
-            | Self::Claude3Sonnet
-            | Self::Claude3_5Sonnet
-            | Self::Claude3_5SonnetV2
-            | Self::Claude3_7Sonnet
-            | Self::Claude3_7SonnetThinking
-            | Self::ClaudeOpus4
-            | Self::ClaudeOpus4Thinking
+            Self::ClaudeHaiku4_5
+            | Self::ClaudeSonnet4
+            | Self::ClaudeSonnet4Thinking
+            | Self::ClaudeSonnet4_5
+            | Self::ClaudeSonnet4_5Thinking
             | Self::ClaudeOpus4_1
             | Self::ClaudeOpus4_1Thinking
             | Self::ClaudeOpus4_5
             | Self::ClaudeOpus4_5Thinking
             | Self::ClaudeOpus4_6
-            | Self::ClaudeOpus4_6Thinking
-            | Self::ClaudeSonnet4
-            | Self::ClaudeSonnet4Thinking
-            | Self::ClaudeSonnet4_5
-            | Self::ClaudeSonnet4_5Thinking
-            | Self::Claude3_5Haiku
-            | Self::ClaudeHaiku4_5 => true,
-
-            // Amazon Nova models (all support tool use)
-            Self::AmazonNovaPremier
-            | Self::AmazonNovaPro
-            | Self::AmazonNovaLite
-            | Self::AmazonNovaMicro => true,
-
-            // AI21 Jamba 1.5 models support tool use
-            Self::AI21Jamba15LargeV1 | Self::AI21Jamba15MiniV1 => true,
-
-            // Cohere Command R models support tool use
-            Self::CohereCommandRV1 | Self::CohereCommandRPlusV1 => true,
-
-            // All other models don't support tool use
-            // Including Meta Llama 3.2, AI21 Jurassic, and others
+            | Self::ClaudeOpus4_6Thinking => true,
+            Self::NovaLite | Self::NovaPro | Self::NovaPremier | Self::Nova2Lite => true,
+            Self::MistralLarge3 | Self::PixtralLarge | Self::MagistralSmall => true,
+            // Gemma accepts toolConfig without error but produces unreliable tool
+            // calls -- malformed JSON args, hallucinated tool names, dropped calls.
+            Self::Qwen3_32B
+            | Self::Qwen3VL235B
+            | Self::Qwen3_235B
+            | Self::Qwen3Next80B
+            | Self::Qwen3Coder30B
+            | Self::Qwen3CoderNext
+            | Self::Qwen3Coder480B => true,
+            Self::MiniMaxM2 => true,
+            Self::KimiK2Thinking | Self::KimiK2_5 => true,
+            Self::DeepSeekR1 | Self::DeepSeekV3_1 | Self::DeepSeekV3_2 => true,
             _ => false,
         }
     }
 
     pub fn supports_images(&self) -> bool {
         match self {
-            // Anthropic Claude 3+ models (all support vision)
-            Self::Claude3Opus
-            | Self::Claude3Sonnet
-            | Self::Claude3_5Sonnet
-            | Self::Claude3_5SonnetV2
-            | Self::Claude3_7Sonnet
-            | Self::Claude3_7SonnetThinking
-            | Self::ClaudeOpus4
-            | Self::ClaudeOpus4Thinking
-            | Self::ClaudeOpus4_1
-            | Self::ClaudeOpus4_1Thinking
-            | Self::ClaudeOpus4_5
-            | Self::ClaudeOpus4_5Thinking
+            Self::ClaudeHaiku4_5
             | Self::ClaudeSonnet4
             | Self::ClaudeSonnet4Thinking
             | Self::ClaudeSonnet4_5
             | Self::ClaudeSonnet4_5Thinking
-            | Self::Claude3_5Haiku
-            | Self::ClaudeHaiku4_5
-            | Self::Claude3Haiku => true,
-
-            // Amazon Nova visual models
-            Self::AmazonNovaPro | Self::AmazonNovaLite => true,
-
-            // Meta Llama 3.2 Vision models
-            Self::MetaLlama3211BInstructV1 | Self::MetaLlama3290BInstructV1 => true,
-
-            // Mistral Pixtral (visual model)
-            Self::MistralPixtralLarge2502V1 => true,
-
-            // Custom models default to no image support
-            Self::Custom { .. } => false,
-
-            // All other models don't support images
+            | Self::ClaudeOpus4_1
+            | Self::ClaudeOpus4_1Thinking
+            | Self::ClaudeOpus4_5
+            | Self::ClaudeOpus4_5Thinking
+            | Self::ClaudeOpus4_6
+            | Self::ClaudeOpus4_6Thinking => true,
+            Self::NovaLite | Self::NovaPro => true,
+            Self::PixtralLarge => true,
+            Self::Qwen3VL235B => true,
+            Self::KimiK2_5 => true,
             _ => false,
         }
     }
 
+    pub fn supports_extended_context(&self) -> bool {
+        matches!(
+            self,
+            Self::ClaudeSonnet4
+                | Self::ClaudeSonnet4Thinking
+                | Self::ClaudeSonnet4_5
+                | Self::ClaudeSonnet4_5Thinking
+                | Self::ClaudeOpus4_5
+                | Self::ClaudeOpus4_5Thinking
+                | Self::ClaudeOpus4_6
+                | Self::ClaudeOpus4_6Thinking
+        )
+    }
+
     pub fn supports_caching(&self) -> bool {
         match self {
-            // Only Claude models on Bedrock support caching
-            // Nova models support only text caching
-            // https://docs.aws.amazon.com/bedrock/latest/userguide/prompt-caching.html#prompt-caching-models
-            Self::Claude3_5Haiku
-            | Self::ClaudeHaiku4_5
-            | Self::Claude3_7Sonnet
-            | Self::Claude3_7SonnetThinking
+            Self::ClaudeHaiku4_5
             | Self::ClaudeSonnet4
             | Self::ClaudeSonnet4Thinking
             | Self::ClaudeSonnet4_5
             | Self::ClaudeSonnet4_5Thinking
-            | Self::ClaudeOpus4
-            | Self::ClaudeOpus4Thinking
             | Self::ClaudeOpus4_1
             | Self::ClaudeOpus4_1Thinking
             | Self::ClaudeOpus4_5
             | Self::ClaudeOpus4_5Thinking
             | Self::ClaudeOpus4_6
             | Self::ClaudeOpus4_6Thinking => true,
-
-            // Custom models - check if they have cache configuration
             Self::Custom {
                 cache_configuration,
                 ..
             } => cache_configuration.is_some(),
-
-            // All other models don't support caching
             _ => false,
         }
     }
 
     pub fn cache_configuration(&self) -> Option<BedrockModelCacheConfiguration> {
         match self {
-            Self::Claude3_7Sonnet
-            | Self::Claude3_7SonnetThinking
-            | Self::ClaudeSonnet4
+            Self::ClaudeSonnet4
             | Self::ClaudeSonnet4Thinking
-            | Self::ClaudeOpus4
-            | Self::ClaudeOpus4Thinking
+            | Self::ClaudeSonnet4_5
+            | Self::ClaudeSonnet4_5Thinking
             | Self::ClaudeOpus4_1
             | Self::ClaudeOpus4_1Thinking
             | Self::ClaudeOpus4_5
@@ -641,97 +535,94 @@ impl Model {
                 max_cache_anchors: 4,
                 min_total_token: 1024,
             }),
-
-            Self::Claude3_5Haiku | Self::ClaudeHaiku4_5 => Some(BedrockModelCacheConfiguration {
+            Self::ClaudeHaiku4_5 => Some(BedrockModelCacheConfiguration {
                 max_cache_anchors: 4,
                 min_total_token: 2048,
             }),
-
             Self::Custom {
                 cache_configuration,
                 ..
             } => cache_configuration.clone(),
-
             _ => None,
         }
     }
 
     pub fn mode(&self) -> BedrockModelMode {
         match self {
-            Model::Claude3_7SonnetThinking => BedrockModelMode::Thinking {
-                budget_tokens: Some(4096),
-            },
-            Model::ClaudeSonnet4Thinking | Model::ClaudeSonnet4_5Thinking => {
+            Self::ClaudeSonnet4Thinking | Self::ClaudeSonnet4_5Thinking => {
                 BedrockModelMode::Thinking {
                     budget_tokens: Some(4096),
                 }
             }
-            Model::ClaudeOpus4Thinking
-            | Model::ClaudeOpus4_1Thinking
-            | Model::ClaudeOpus4_5Thinking => BedrockModelMode::Thinking {
-                budget_tokens: Some(4096),
-            },
-            Model::ClaudeOpus4_6Thinking => BedrockModelMode::AdaptiveThinking {
+            Self::ClaudeOpus4_1Thinking | Self::ClaudeOpus4_5Thinking => {
+                BedrockModelMode::Thinking {
+                    budget_tokens: Some(4096),
+                }
+            }
+            Self::ClaudeOpus4_6Thinking => BedrockModelMode::AdaptiveThinking {
                 effort: BedrockAdaptiveThinkingEffort::default(),
             },
             _ => BedrockModelMode::Default,
         }
     }
 
-    pub fn supports_extended_context(&self) -> bool {
-        matches!(
-            self,
-            Model::ClaudeSonnet4
-                | Model::ClaudeSonnet4Thinking
-                | Model::ClaudeSonnet4_5
-                | Model::ClaudeSonnet4_5Thinking
-                | Model::ClaudeOpus4_5
-                | Model::ClaudeOpus4_5Thinking
-                | Model::ClaudeOpus4_6
-                | Model::ClaudeOpus4_6Thinking
-        )
-    }
-
     pub fn cross_region_inference_id(
         &self,
         region: &str,
         allow_global: bool,
     ) -> anyhow::Result<String> {
-        // List derived from here:
-        // https://docs.aws.amazon.com/bedrock/latest/userguide/inference-profiles-support.html#inference-profiles-support-system
         let model_id = self.request_id();
 
         let supports_global = matches!(
             self,
-            Model::ClaudeOpus4_5
-                | Model::ClaudeOpus4_5Thinking
-                | Model::ClaudeOpus4_6
-                | Model::ClaudeOpus4_6Thinking
-                | Model::ClaudeHaiku4_5
-                | Model::ClaudeSonnet4
-                | Model::ClaudeSonnet4Thinking
-                | Model::ClaudeSonnet4_5
-                | Model::ClaudeSonnet4_5Thinking
+            Self::ClaudeHaiku4_5
+                | Self::ClaudeSonnet4
+                | Self::ClaudeSonnet4Thinking
+                | Self::ClaudeSonnet4_5
+                | Self::ClaudeSonnet4_5Thinking
+                | Self::ClaudeOpus4_5
+                | Self::ClaudeOpus4_5Thinking
+                | Self::ClaudeOpus4_6
+                | Self::ClaudeOpus4_6Thinking
+                | Self::Nova2Lite
         );
 
+        // Determine region group based on AWS region
         let region_group = if region.starts_with("us-gov-") {
             "us-gov"
-        } else if region.starts_with("us-")
-            || region.starts_with("ca-")
-            || region.starts_with("sa-")
-        {
+        } else if region.starts_with("us-") || region.starts_with("sa-") {
             if allow_global && supports_global {
                 "global"
             } else {
                 "us"
             }
+        } else if region.starts_with("ca-") {
+            if allow_global && supports_global {
+                "global"
+            } else {
+                "ca"
+            }
         } else if region.starts_with("eu-") {
             if allow_global && supports_global {
                 "global"
             } else {
                 "eu"
             }
-        } else if region.starts_with("ap-") || region == "me-central-1" || region == "me-south-1" {
+        } else if region == "ap-southeast-2" || region == "ap-southeast-4" {
+            // Australia
+            if allow_global && supports_global {
+                "global"
+            } else {
+                "au"
+            }
+        } else if region == "ap-northeast-1" || region == "ap-northeast-3" {
+            // Japan
+            if allow_global && supports_global {
+                "global"
+            } else {
+                "jp"
+            }
+        } else if region.starts_with("ap-") || region.starts_with("me-") {
             if allow_global && supports_global {
                 "global"
             } else {
@@ -741,134 +632,102 @@ impl Model {
             anyhow::bail!("Unsupported Region {region}");
         };
 
-        match (self, region_group, region) {
-            (Model::Custom { .. }, _, _) => Ok(self.request_id().into()),
+        match (self, region_group) {
+            (Self::Custom { .. }, _) => Ok(model_id.into()),
 
+            // Global inference profiles
             (
-                Model::ClaudeOpus4_5
-                | Model::ClaudeOpus4_5Thinking
-                | Model::ClaudeOpus4_6
-                | Model::ClaudeOpus4_6Thinking
-                | Model::ClaudeHaiku4_5
-                | Model::ClaudeSonnet4
-                | Model::ClaudeSonnet4Thinking
-                | Model::ClaudeSonnet4_5
-                | Model::ClaudeSonnet4_5Thinking,
+                Self::ClaudeHaiku4_5
+                | Self::ClaudeSonnet4
+                | Self::ClaudeSonnet4Thinking
+                | Self::ClaudeSonnet4_5
+                | Self::ClaudeSonnet4_5Thinking
+                | Self::ClaudeOpus4_5
+                | Self::ClaudeOpus4_5Thinking
+                | Self::ClaudeOpus4_6
+                | Self::ClaudeOpus4_6Thinking
+                | Self::Nova2Lite,
                 "global",
-                _,
             ) => Ok(format!("{}.{}", region_group, model_id)),
 
+            // US Government region inference profiles
+            (Self::ClaudeSonnet4_5 | Self::ClaudeSonnet4_5Thinking, "us-gov") => {
+                Ok(format!("{}.{}", region_group, model_id))
+            }
+
+            // US region inference profiles
             (
-                Model::Claude3Haiku
-                | Model::Claude3_5Sonnet
-                | Model::Claude3_7Sonnet
-                | Model::Claude3_7SonnetThinking
-                | Model::ClaudeSonnet4_5
-                | Model::ClaudeSonnet4_5Thinking,
-                "us-gov",
-                _,
+                Self::ClaudeHaiku4_5
+                | Self::ClaudeSonnet4
+                | Self::ClaudeSonnet4Thinking
+                | Self::ClaudeSonnet4_5
+                | Self::ClaudeSonnet4_5Thinking
+                | Self::ClaudeOpus4_1
+                | Self::ClaudeOpus4_1Thinking
+                | Self::ClaudeOpus4_5
+                | Self::ClaudeOpus4_5Thinking
+                | Self::ClaudeOpus4_6
+                | Self::ClaudeOpus4_6Thinking
+                | Self::Llama4Scout17B
+                | Self::Llama4Maverick17B
+                | Self::NovaLite
+                | Self::NovaPro
+                | Self::NovaPremier
+                | Self::Nova2Lite
+                | Self::PixtralLarge
+                | Self::DeepSeekR1,
+                "us",
             ) => Ok(format!("{}.{}", region_group, model_id)),
 
-            (
-                Model::ClaudeHaiku4_5 | Model::ClaudeSonnet4_5 | Model::ClaudeSonnet4_5Thinking,
-                "apac",
-                "ap-southeast-2" | "ap-southeast-4",
-            ) => Ok(format!("au.{}", model_id)),
+            // Canada region inference profiles
+            (Self::NovaLite, "ca") => Ok(format!("{}.{}", region_group, model_id)),
 
+            // EU region inference profiles
             (
-                Model::ClaudeHaiku4_5 | Model::ClaudeSonnet4_5 | Model::ClaudeSonnet4_5Thinking,
-                "apac",
-                "ap-northeast-1" | "ap-northeast-3",
-            ) => Ok(format!("jp.{}", model_id)),
-
-            (Model::AmazonNovaLite, "us", r) if r.starts_with("ca-") => {
-                Ok(format!("ca.{}", model_id))
-            }
+                Self::ClaudeHaiku4_5
+                | Self::ClaudeSonnet4
+                | Self::ClaudeSonnet4_5
+                | Self::ClaudeSonnet4_5Thinking
+                | Self::ClaudeOpus4_6
+                | Self::ClaudeOpus4_6Thinking
+                | Self::NovaLite
+                | Self::NovaPro
+                | Self::Nova2Lite,
+                "eu",
+            ) => Ok(format!("{}.{}", region_group, model_id)),
 
+            // Australia region inference profiles
             (
-                Model::AmazonNovaPremier
-                | Model::AmazonNovaLite
-                | Model::AmazonNovaMicro
-                | Model::AmazonNovaPro
-                | Model::Claude3_5Haiku
-                | Model::ClaudeHaiku4_5
-                | Model::Claude3_5Sonnet
-                | Model::Claude3_5SonnetV2
-                | Model::Claude3_7Sonnet
-                | Model::Claude3_7SonnetThinking
-                | Model::ClaudeSonnet4
-                | Model::ClaudeSonnet4Thinking
-                | Model::ClaudeSonnet4_5
-                | Model::ClaudeSonnet4_5Thinking
-                | Model::ClaudeOpus4
-                | Model::ClaudeOpus4Thinking
-                | Model::ClaudeOpus4_1
-                | Model::ClaudeOpus4_1Thinking
-                | Model::ClaudeOpus4_5
-                | Model::ClaudeOpus4_5Thinking
-                | Model::ClaudeOpus4_6
-                | Model::ClaudeOpus4_6Thinking
-                | Model::Claude3Haiku
-                | Model::Claude3Opus
-                | Model::Claude3Sonnet
-                | Model::DeepSeekR1
-                | Model::MetaLlama31405BInstructV1
-                | Model::MetaLlama3170BInstructV1_128k
-                | Model::MetaLlama3170BInstructV1
-                | Model::MetaLlama318BInstructV1_128k
-                | Model::MetaLlama318BInstructV1
-                | Model::MetaLlama3211BInstructV1
-                | Model::MetaLlama321BInstructV1
-                | Model::MetaLlama323BInstructV1
-                | Model::MetaLlama3290BInstructV1
-                | Model::MetaLlama3370BInstructV1
-                | Model::MetaLlama4Maverick17BInstructV1
-                | Model::MetaLlama4Scout17BInstructV1
-                | Model::MistralPixtralLarge2502V1
-                | Model::PalmyraWriterX4
-                | Model::PalmyraWriterX5,
-                "us",
-                _,
+                Self::ClaudeHaiku4_5
+                | Self::ClaudeSonnet4_5
+                | Self::ClaudeSonnet4_5Thinking
+                | Self::ClaudeOpus4_6
+                | Self::ClaudeOpus4_6Thinking,
+                "au",
             ) => Ok(format!("{}.{}", region_group, model_id)),
 
+            // Japan region inference profiles
             (
-                Model::AmazonNovaLite
-                | Model::AmazonNovaMicro
-                | Model::AmazonNovaPro
-                | Model::Claude3_5Sonnet
-                | Model::ClaudeHaiku4_5
-                | Model::Claude3_7Sonnet
-                | Model::Claude3_7SonnetThinking
-                | Model::ClaudeSonnet4
-                | Model::ClaudeSonnet4_5
-                | Model::ClaudeSonnet4_5Thinking
-                | Model::ClaudeOpus4_6
-                | Model::ClaudeOpus4_6Thinking
-                | Model::Claude3Haiku
-                | Model::Claude3Sonnet
-                | Model::MetaLlama321BInstructV1
-                | Model::MetaLlama323BInstructV1
-                | Model::MistralPixtralLarge2502V1,
-                "eu",
-                _,
+                Self::ClaudeHaiku4_5
+                | Self::ClaudeSonnet4_5
+                | Self::ClaudeSonnet4_5Thinking
+                | Self::Nova2Lite,
+                "jp",
             ) => Ok(format!("{}.{}", region_group, model_id)),
 
+            // APAC region inference profiles (other than AU/JP)
             (
-                Model::AmazonNovaLite
-                | Model::AmazonNovaMicro
-                | Model::AmazonNovaPro
-                | Model::Claude3_5Sonnet
-                | Model::Claude3_5SonnetV2
-                | Model::ClaudeHaiku4_5
-                | Model::Claude3_7Sonnet
-                | Model::Claude3_7SonnetThinking
-                | Model::ClaudeSonnet4
-                | Model::Claude3Haiku
-                | Model::Claude3Sonnet,
+                Self::ClaudeHaiku4_5
+                | Self::ClaudeSonnet4
+                | Self::ClaudeSonnet4_5
+                | Self::ClaudeSonnet4_5Thinking
+                | Self::NovaLite
+                | Self::NovaPro
+                | Self::Nova2Lite,
                 "apac",
-                _,
             ) => Ok(format!("{}.{}", region_group, model_id)),
 
+            // Default: use model ID directly
             _ => Ok(model_id.into()),
         }
     }

crates/language_models/src/provider/bedrock.rs

@@ -600,18 +600,19 @@ impl BedrockModel {
         cx: &AsyncApp,
     ) -> BoxFuture<
         'static,
-        Result<BoxStream<'static, Result<BedrockStreamingResponse, BedrockError>>>,
+        Result<BoxStream<'static, Result<BedrockStreamingResponse, anyhow::Error>>, BedrockError>,
     > {
         let Ok(runtime_client) = self
             .get_or_init_client(cx)
             .cloned()
             .context("Bedrock client not initialized")
         else {
-            return futures::future::ready(Err(anyhow!("App state dropped"))).boxed();
+            return futures::future::ready(Err(BedrockError::Other(anyhow!("App state dropped"))))
+                .boxed();
         };
 
         let task = Tokio::spawn(cx, bedrock::stream_completion(runtime_client, request));
-        async move { task.await.map_err(|err| anyhow!(err))? }.boxed()
+        async move { task.await.map_err(|e| BedrockError::Other(e.into()))? }.boxed()
     }
 }
 
@@ -715,6 +716,7 @@ impl LanguageModel for BedrockModel {
             self.model.max_output_tokens(),
             self.model.mode(),
             self.model.supports_caching(),
+            self.model.supports_tool_use(),
             use_extended_context,
         ) {
             Ok(request) => request,
@@ -722,8 +724,44 @@ impl LanguageModel for BedrockModel {
         };
 
         let request = self.stream_completion(request, cx);
+        let display_name = self.model.display_name().to_string();
         let future = self.request_limiter.stream(async move {
-            let response = request.await.map_err(|err| anyhow!(err))?;
+            let response = request.await.map_err(|err| match err {
+                BedrockError::Validation(ref msg) => {
+                    if msg.contains("model identifier is invalid") {
+                        LanguageModelCompletionError::Other(anyhow!(
+                            "{display_name} is not available in {region}. \
+                                 Try switching to a region where this model is supported."
+                        ))
+                    } else {
+                        LanguageModelCompletionError::BadRequestFormat {
+                            provider: PROVIDER_NAME,
+                            message: msg.clone(),
+                        }
+                    }
+                }
+                BedrockError::RateLimited => LanguageModelCompletionError::RateLimitExceeded {
+                    provider: PROVIDER_NAME,
+                    retry_after: None,
+                },
+                BedrockError::ServiceUnavailable => {
+                    LanguageModelCompletionError::ServerOverloaded {
+                        provider: PROVIDER_NAME,
+                        retry_after: None,
+                    }
+                }
+                BedrockError::AccessDenied(msg) => LanguageModelCompletionError::PermissionError {
+                    provider: PROVIDER_NAME,
+                    message: msg,
+                },
+                BedrockError::InternalServer(msg) => {
+                    LanguageModelCompletionError::ApiInternalServerError {
+                        provider: PROVIDER_NAME,
+                        message: msg,
+                    }
+                }
+                other => LanguageModelCompletionError::Other(anyhow!(other)),
+            })?;
             let events = map_to_language_model_completion_events(response);
 
             if deny_tool_calls {
@@ -771,6 +809,7 @@ pub fn into_bedrock(
     max_output_tokens: u64,
     mode: BedrockModelMode,
     supports_caching: bool,
+    supports_tool_use: bool,
     allow_extended_context: bool,
 ) -> Result<bedrock::Request> {
     let mut new_messages: Vec<BedrockMessage> = Vec::new();
@@ -965,28 +1004,32 @@ pub fn into_bedrock(
         }
     }
 
-    let mut tool_spec: Vec<BedrockTool> = request
-        .tools
-        .iter()
-        .filter_map(|tool| {
-            Some(BedrockTool::ToolSpec(
-                BedrockToolSpec::builder()
-                    .name(tool.name.clone())
-                    .description(tool.description.clone())
-                    .input_schema(BedrockToolInputSchema::Json(value_to_aws_document(
-                        &tool.input_schema,
-                    )))
-                    .build()
-                    .log_err()?,
-            ))
-        })
-        .collect();
+    let mut tool_spec: Vec<BedrockTool> = if supports_tool_use {
+        request
+            .tools
+            .iter()
+            .filter_map(|tool| {
+                Some(BedrockTool::ToolSpec(
+                    BedrockToolSpec::builder()
+                        .name(tool.name.clone())
+                        .description(tool.description.clone())
+                        .input_schema(BedrockToolInputSchema::Json(value_to_aws_document(
+                            &tool.input_schema,
+                        )))
+                        .build()
+                        .log_err()?,
+                ))
+            })
+            .collect()
+    } else {
+        Vec::new()
+    };
 
     // Bedrock requires toolConfig when messages contain tool use/result blocks.
     // If no tools are defined but messages contain tool content (e.g., when
     // summarising a conversation that used tools), add a dummy tool to satisfy
     // the API requirement.
-    if tool_spec.is_empty() && messages_contain_tool_content {
+    if supports_tool_use && tool_spec.is_empty() && messages_contain_tool_content {
         tool_spec.push(BedrockTool::ToolSpec(
             BedrockToolSpec::builder()
                 .name("_placeholder")
@@ -1020,17 +1063,23 @@ pub fn into_bedrock(
             BedrockToolChoice::Auto(BedrockAutoToolChoice::builder().build())
         }
     };
-    let tool_config: BedrockToolConfig = BedrockToolConfig::builder()
-        .set_tools(Some(tool_spec))
-        .tool_choice(tool_choice)
-        .build()?;
+    let tool_config = if tool_spec.is_empty() {
+        None
+    } else {
+        Some(
+            BedrockToolConfig::builder()
+                .set_tools(Some(tool_spec))
+                .tool_choice(tool_choice)
+                .build()?,
+        )
+    };
 
     Ok(bedrock::Request {
         model,
         messages: new_messages,
         max_tokens: max_output_tokens,
         system: Some(system_message),
-        tools: Some(tool_config),
+        tools: tool_config,
         thinking: if request.thinking_allowed {
             match mode {
                 BedrockModelMode::Thinking { budget_tokens } => {
@@ -1116,7 +1165,7 @@ pub fn get_bedrock_tokens(
 }
 
 pub fn map_to_language_model_completion_events(
-    events: Pin<Box<dyn Send + Stream<Item = Result<BedrockStreamingResponse, BedrockError>>>>,
+    events: Pin<Box<dyn Send + Stream<Item = Result<BedrockStreamingResponse, anyhow::Error>>>>,
 ) -> impl Stream<Item = Result<LanguageModelCompletionEvent, LanguageModelCompletionError>> {
     struct RawToolUse {
         id: String,
@@ -1125,13 +1174,15 @@ pub fn map_to_language_model_completion_events(
     }
 
     struct State {
-        events: Pin<Box<dyn Send + Stream<Item = Result<BedrockStreamingResponse, BedrockError>>>>,
+        events: Pin<Box<dyn Send + Stream<Item = Result<BedrockStreamingResponse, anyhow::Error>>>>,
         tool_uses_by_index: HashMap<i32, RawToolUse>,
+        emitted_tool_use: bool,
     }
 
     let initial_state = State {
         events,
         tool_uses_by_index: HashMap::default(),
+        emitted_tool_use: false,
     };
 
     futures::stream::unfold(initial_state, |mut state| async move {
@@ -1190,10 +1241,13 @@ pub fn map_to_language_model_completion_events(
                             }
                             None
                         }
+                        ConverseStreamOutput::MessageStart(_) => None,
                         ConverseStreamOutput::ContentBlockStop(cb_stop) => state
                             .tool_uses_by_index
                             .remove(&cb_stop.content_block_index)
                             .map(|tool_use| {
+                                state.emitted_tool_use = true;
+
                                 let input = parse_tool_arguments(&tool_use.input_json)
                                     .unwrap_or_else(|_| Value::Object(Default::default()));
 
@@ -1223,9 +1277,16 @@ pub fn map_to_language_model_completion_events(
                             }))
                         }),
                         ConverseStreamOutput::MessageStop(message_stop) => {
-                            let stop_reason = match message_stop.stop_reason {
-                                StopReason::ToolUse => language_model::StopReason::ToolUse,
-                                _ => language_model::StopReason::EndTurn,
+                            let stop_reason = if state.emitted_tool_use {
+                                // Some models (e.g. Kimi) send EndTurn even when
+                                // they've made tool calls. Trust the content over
+                                // the stop reason.
+                                language_model::StopReason::ToolUse
+                            } else {
+                                match message_stop.stop_reason {
+                                    StopReason::ToolUse => language_model::StopReason::ToolUse,
+                                    _ => language_model::StopReason::EndTurn,
+                                }
                             };
                             Some(Ok(LanguageModelCompletionEvent::Stop(stop_reason)))
                         }