diff --git a/internal/providers/configs/chutes.json b/internal/providers/configs/chutes.json index 5b8253d2636fc5ff024339e6884c4976580d76f9..71e79bd900bbdaa986ef20b2f97c6886f22d199f 100644 --- a/internal/providers/configs/chutes.json +++ b/internal/providers/configs/chutes.json @@ -103,24 +103,6 @@ "default_reasoning_effort": "medium", "supports_attachments": false }, - { - "id": "deepseek-ai/DeepSeek-V3.1-Terminus-TEE", - "name": "DeepSeek-V3.1-Terminus-TEE", - "cost_per_1m_in": 0.27, - "cost_per_1m_out": 1, - "cost_per_1m_in_cached": 0.135, - "cost_per_1m_out_cached": 0, - "context_window": 163840, - "default_max_tokens": 65536, - "can_reason": true, - "reasoning_levels": [ - "low", - "medium", - "high" - ], - "default_reasoning_effort": "medium", - "supports_attachments": false - }, { "id": "deepseek-ai/DeepSeek-V3.2-TEE", "name": "DeepSeek-V3.2-TEE", @@ -139,42 +121,6 @@ "default_reasoning_effort": "medium", "supports_attachments": false }, - { - "id": "zai-org/GLM-4.6-FP8", - "name": "GLM-4.6-FP8", - "cost_per_1m_in": 0.2989, - "cost_per_1m_out": 1.1957, - "cost_per_1m_in_cached": 0.14945, - "cost_per_1m_out_cached": 0, - "context_window": 202752, - "default_max_tokens": 65535, - "can_reason": true, - "reasoning_levels": [ - "low", - "medium", - "high" - ], - "default_reasoning_effort": "medium", - "supports_attachments": false - }, - { - "id": "zai-org/GLM-4.6-TEE", - "name": "GLM-4.6-TEE", - "cost_per_1m_in": 0.4, - "cost_per_1m_out": 1.7, - "cost_per_1m_in_cached": 0.2, - "cost_per_1m_out_cached": 0, - "context_window": 202752, - "default_max_tokens": 65536, - "can_reason": true, - "reasoning_levels": [ - "low", - "medium", - "high" - ], - "default_reasoning_effort": "medium", - "supports_attachments": false - }, { "id": "zai-org/GLM-4.6V", "name": "GLM-4.6V", diff --git a/internal/providers/configs/cortecs.json b/internal/providers/configs/cortecs.json index 75c4ff1cc6e91a1611db30631b3e9ba34c781fb6..41d341e53a4b31700241832d64398bb88b60e3c7 100644 --- a/internal/providers/configs/cortecs.json +++ b/internal/providers/configs/cortecs.json @@ -25,6 +25,24 @@ "default_reasoning_effort": "medium", "supports_attachments": false }, + { + "id": "qwen3.5-122b-a10b", + "name": "Qwen3.5 122B A10B", + "cost_per_1m_in": 0.466, + "cost_per_1m_out": 3.261, + "cost_per_1m_in_cached": 0, + "cost_per_1m_out_cached": 0, + "context_window": 262144, + "default_max_tokens": 26214, + "can_reason": true, + "reasoning_levels": [ + "low", + "medium", + "high" + ], + "default_reasoning_effort": "medium", + "supports_attachments": false + }, { "id": "qwen3.5-9b", "name": "Qwen3.5 9B", diff --git a/internal/providers/configs/huggingface.json b/internal/providers/configs/huggingface.json index 8261c3998ee8cfc2d3678d660bb01af338a5fda4..8538eeace5cc6217656fea1f526a17b145be877f 100644 --- a/internal/providers/configs/huggingface.json +++ b/internal/providers/configs/huggingface.json @@ -175,6 +175,18 @@ "can_reason": false, "supports_attachments": false }, + { + "id": "openai/gpt-oss-120b:cerebras", + "name": "openai/gpt-oss-120b (cerebras)", + "cost_per_1m_in": 0.25, + "cost_per_1m_out": 0.69, + "cost_per_1m_in_cached": 0, + "cost_per_1m_out_cached": 0, + "context_window": 131072, + "default_max_tokens": 8192, + "can_reason": false, + "supports_attachments": false + }, { "id": "openai/gpt-oss-120b:fireworks-ai", "name": "openai/gpt-oss-120b (fireworks-ai)", diff --git a/internal/providers/configs/openrouter.json b/internal/providers/configs/openrouter.json index 3d3b514131c0fc94e86d41f24cf26f1b52c97201..5727fc95ffe1477ce332e2d75a9abd70ca52233d 100644 --- a/internal/providers/configs/openrouter.json +++ b/internal/providers/configs/openrouter.json @@ -532,12 +532,12 @@ { "id": "deepseek/deepseek-chat-v3.1", "name": "DeepSeek: DeepSeek V3.1", - "cost_per_1m_in": 0.6, - "cost_per_1m_out": 1.7, + "cost_per_1m_in": 0.27, + "cost_per_1m_out": 1, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.135, "context_window": 163840, - "default_max_tokens": 16384, + "default_max_tokens": 32768, "can_reason": true, "reasoning_levels": [ "low", @@ -550,12 +550,12 @@ { "id": "deepseek/deepseek-v3.1-terminus", "name": "DeepSeek: DeepSeek V3.1 Terminus", - "cost_per_1m_in": 0.27, - "cost_per_1m_out": 1, + "cost_per_1m_in": 0.21, + "cost_per_1m_out": 0.79, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.135, + "cost_per_1m_out_cached": 0.13, "context_window": 163840, - "default_max_tokens": 32768, + "default_max_tokens": 16384, "can_reason": true, "reasoning_levels": [ "low", @@ -622,12 +622,12 @@ { "id": "deepseek/deepseek-r1-0528", "name": "DeepSeek: R1 0528", - "cost_per_1m_in": 0.8, - "cost_per_1m_out": 2.4, + "cost_per_1m_in": 0.45, + "cost_per_1m_out": 2.15, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.225, "context_window": 163840, - "default_max_tokens": 16384, + "default_max_tokens": 32768, "can_reason": true, "reasoning_levels": [ "low", @@ -789,7 +789,7 @@ "cost_per_1m_in_cached": 0.08333, "cost_per_1m_out_cached": 0.05, "context_window": 1048576, - "default_max_tokens": 32768, + "default_max_tokens": 32767, "can_reason": true, "reasoning_levels": [ "low", @@ -999,19 +999,19 @@ "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 13107, + "default_max_tokens": 8192, "can_reason": false, "supports_attachments": false }, { "id": "meta-llama/llama-3.3-70b-instruct", "name": "Meta: Llama 3.3 70B Instruct", - "cost_per_1m_in": 0.25, - "cost_per_1m_out": 0.75, + "cost_per_1m_in": 0.135, + "cost_per_1m_out": 0.4, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 13107, + "default_max_tokens": 60000, "can_reason": false, "supports_attachments": false }, @@ -1051,15 +1051,33 @@ "can_reason": false, "supports_attachments": true }, + { + "id": "minimax/minimax-m1", + "name": "MiniMax: MiniMax M1", + "cost_per_1m_in": 0.44, + "cost_per_1m_out": 1.76, + "cost_per_1m_in_cached": 0, + "cost_per_1m_out_cached": 0, + "context_window": 1000000, + "default_max_tokens": 20000, + "can_reason": true, + "reasoning_levels": [ + "low", + "medium", + "high" + ], + "default_reasoning_effort": "medium", + "supports_attachments": false + }, { "id": "minimax/minimax-m2", "name": "MiniMax: MiniMax M2", "cost_per_1m_in": 0.255, - "cost_per_1m_out": 1, + "cost_per_1m_out": 1.02, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.03, - "context_window": 196608, - "default_max_tokens": 98304, + "cost_per_1m_out_cached": 0, + "context_window": 204800, + "default_max_tokens": 65536, "can_reason": true, "reasoning_levels": [ "low", @@ -1095,7 +1113,7 @@ "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0.03, "context_window": 204800, - "default_max_tokens": 65536, + "default_max_tokens": 65550, "can_reason": true, "reasoning_levels": [ "low", @@ -1228,36 +1246,36 @@ { "id": "mistralai/ministral-14b-2512", "name": "Mistral: Ministral 3 14B 2512", - "cost_per_1m_in": 0.35, - "cost_per_1m_out": 0.35, + "cost_per_1m_in": 0.2, + "cost_per_1m_out": 0.2, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.02, "context_window": 262144, - "default_max_tokens": 131072, + "default_max_tokens": 26214, "can_reason": false, "supports_attachments": true }, { "id": "mistralai/ministral-3b-2512", "name": "Mistral: Ministral 3 3B 2512", - "cost_per_1m_in": 0.35, - "cost_per_1m_out": 0.35, + "cost_per_1m_in": 0.1, + "cost_per_1m_out": 0.1, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.01, "context_window": 131072, - "default_max_tokens": 65536, + "default_max_tokens": 13107, "can_reason": false, "supports_attachments": true }, { "id": "mistralai/ministral-8b-2512", "name": "Mistral: Ministral 3 8B 2512", - "cost_per_1m_in": 0.15, - "cost_per_1m_out": 0.15, + "cost_per_1m_in": 0.3, + "cost_per_1m_out": 0.3, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.015, + "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 26214, + "default_max_tokens": 131072, "can_reason": false, "supports_attachments": true }, @@ -1429,9 +1447,9 @@ "cost_per_1m_in": 0.6, "cost_per_1m_out": 2.5, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.6, + "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 16384, + "default_max_tokens": 131072, "can_reason": false, "supports_attachments": false }, @@ -1537,24 +1555,6 @@ "default_reasoning_effort": "medium", "supports_attachments": false }, - { - "id": "nvidia/nemotron-3-super-120b-a12b", - "name": "NVIDIA: Nemotron 3 Super", - "cost_per_1m_in": 0.1, - "cost_per_1m_out": 0.5, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.1, - "context_window": 262144, - "default_max_tokens": 26214, - "can_reason": true, - "reasoning_levels": [ - "low", - "medium", - "high" - ], - "default_reasoning_effort": "medium", - "supports_attachments": false - }, { "id": "nvidia/nemotron-3-super-120b-a12b:free", "name": "NVIDIA: Nemotron 3 Super (free)", @@ -1707,7 +1707,7 @@ "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0.5, "context_window": 1047576, - "default_max_tokens": 104757, + "default_max_tokens": 16384, "can_reason": false, "supports_attachments": true }, @@ -1963,7 +1963,7 @@ "cost_per_1m_in": 1.25, "cost_per_1m_out": 10, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.125, + "cost_per_1m_out_cached": 0.13, "context_window": 400000, "default_max_tokens": 64000, "can_reason": true, @@ -2067,7 +2067,7 @@ "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0.175, "context_window": 128000, - "default_max_tokens": 8192, + "default_max_tokens": 16000, "can_reason": false, "supports_attachments": true }, @@ -2212,12 +2212,12 @@ { "id": "openai/gpt-oss-120b", "name": "OpenAI: gpt-oss-120b", - "cost_per_1m_in": 0.039, - "cost_per_1m_out": 0.19, + "cost_per_1m_in": 0.05, + "cost_per_1m_out": 0.25, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 13107, + "default_max_tokens": 16384, "can_reason": true, "reasoning_levels": [ "low", @@ -2620,12 +2620,12 @@ { "id": "qwen/qwen3-235b-a22b-2507", "name": "Qwen: Qwen3 235B A22B Instruct 2507", - "cost_per_1m_in": 0.22, - "cost_per_1m_out": 0.88, + "cost_per_1m_in": 0.2, + "cost_per_1m_out": 0.8, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 8192, + "default_max_tokens": 131072, "can_reason": false, "supports_attachments": false }, @@ -2650,12 +2650,12 @@ { "id": "qwen/qwen3-30b-a3b", "name": "Qwen: Qwen3 30B A3B", - "cost_per_1m_in": 0.15, - "cost_per_1m_out": 0.6, + "cost_per_1m_in": 0.13, + "cost_per_1m_out": 0.52, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.15, + "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 65536, + "default_max_tokens": 4096, "can_reason": true, "reasoning_levels": [ "low", @@ -2680,11 +2680,11 @@ { "id": "qwen/qwen3-30b-a3b-thinking-2507", "name": "Qwen: Qwen3 30B A3B Thinking 2507", - "cost_per_1m_in": 0.08, - "cost_per_1m_out": 0.4, + "cost_per_1m_in": 0.09, + "cost_per_1m_out": 0.3, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.08, - "context_window": 131072, + "cost_per_1m_out_cached": 0, + "context_window": 262144, "default_max_tokens": 65536, "can_reason": true, "reasoning_levels": [ @@ -2746,8 +2746,8 @@ { "id": "qwen/qwen3-coder", "name": "Qwen: Qwen3 Coder 480B A35B", - "cost_per_1m_in": 0.975, - "cost_per_1m_out": 4.875, + "cost_per_1m_in": 0.22, + "cost_per_1m_out": 1.8, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 262144, @@ -2782,12 +2782,12 @@ { "id": "qwen/qwen3-coder-next", "name": "Qwen: Qwen3 Coder Next", - "cost_per_1m_in": 0.15, - "cost_per_1m_out": 0.8, + "cost_per_1m_in": 0.12, + "cost_per_1m_out": 0.75, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.12, + "cost_per_1m_out_cached": 0.06, "context_window": 262144, - "default_max_tokens": 131072, + "default_max_tokens": 32768, "can_reason": false, "supports_attachments": false }, @@ -2878,12 +2878,12 @@ { "id": "qwen/qwen3-vl-235b-a22b-instruct", "name": "Qwen: Qwen3 VL 235B A22B Instruct", - "cost_per_1m_in": 0.21, - "cost_per_1m_out": 1.9, + "cost_per_1m_in": 0.3, + "cost_per_1m_out": 1.5, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, - "context_window": 131072, - "default_max_tokens": 16384, + "context_window": 262144, + "default_max_tokens": 131072, "can_reason": false, "supports_attachments": true }, @@ -2908,20 +2908,20 @@ { "id": "qwen/qwen3-vl-30b-a3b-instruct", "name": "Qwen: Qwen3 VL 30B A3B Instruct", - "cost_per_1m_in": 0.29, - "cost_per_1m_out": 1, + "cost_per_1m_in": 0.13, + "cost_per_1m_out": 0.52, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, - "context_window": 262144, - "default_max_tokens": 131072, + "context_window": 131072, + "default_max_tokens": 16384, "can_reason": false, "supports_attachments": true }, { "id": "qwen/qwen3-vl-30b-a3b-thinking", "name": "Qwen: Qwen3 VL 30B A3B Thinking", - "cost_per_1m_in": 0.13, - "cost_per_1m_out": 1.56, + "cost_per_1m_in": 0.2, + "cost_per_1m_out": 1, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, @@ -2985,7 +2985,7 @@ "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 131072, + "default_max_tokens": 32768, "can_reason": true, "reasoning_levels": [ "low", @@ -3016,10 +3016,10 @@ { "id": "qwen/qwen3.5-122b-a10b", "name": "Qwen: Qwen3.5-122B-A10B", - "cost_per_1m_in": 0.26, - "cost_per_1m_out": 2.08, + "cost_per_1m_in": 0.3, + "cost_per_1m_out": 2.4, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.3, "context_window": 262144, "default_max_tokens": 32768, "can_reason": true, @@ -3034,8 +3034,8 @@ { "id": "qwen/qwen3.5-27b", "name": "Qwen: Qwen3.5-27B", - "cost_per_1m_in": 0.195, - "cost_per_1m_out": 1.56, + "cost_per_1m_in": 0.3, + "cost_per_1m_out": 2.4, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 262144, @@ -3052,8 +3052,8 @@ { "id": "qwen/qwen3.5-35b-a3b", "name": "Qwen: Qwen3.5-35B-A3B", - "cost_per_1m_in": 0.25, - "cost_per_1m_out": 1, + "cost_per_1m_in": 0.23, + "cost_per_1m_out": 1.8, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 262144, @@ -3139,9 +3139,9 @@ "cost_per_1m_in": 0.1, "cost_per_1m_out": 0.3, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.02, - "context_window": 256000, - "default_max_tokens": 128000, + "cost_per_1m_out_cached": 0, + "context_window": 262144, + "default_max_tokens": 32768, "can_reason": true, "reasoning_levels": [ "low", @@ -3235,9 +3235,9 @@ "cost_per_1m_in": 0.1, "cost_per_1m_out": 0.3, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.02, + "cost_per_1m_out_cached": 0.01, "context_window": 262144, - "default_max_tokens": 16000, + "default_max_tokens": 32768, "can_reason": true, "reasoning_levels": [ "low", @@ -3301,9 +3301,9 @@ "cost_per_1m_in": 0.6, "cost_per_1m_out": 2.2, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.11, "context_window": 131072, - "default_max_tokens": 13107, + "default_max_tokens": 49152, "can_reason": true, "reasoning_levels": [ "low", @@ -3514,10 +3514,10 @@ { "id": "x-ai/grok-3", "name": "xAI: Grok 3", - "cost_per_1m_in": 5, - "cost_per_1m_out": 25, + "cost_per_1m_in": 3, + "cost_per_1m_out": 15, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 1.25, + "cost_per_1m_out_cached": 0.75, "context_window": 131072, "default_max_tokens": 13107, "can_reason": false, @@ -3526,10 +3526,10 @@ { "id": "x-ai/grok-3-beta", "name": "xAI: Grok 3 Beta", - "cost_per_1m_in": 5, - "cost_per_1m_out": 25, + "cost_per_1m_in": 3, + "cost_per_1m_out": 15, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 1.25, + "cost_per_1m_out_cached": 0.75, "context_window": 131072, "default_max_tokens": 13107, "can_reason": false, diff --git a/internal/providers/configs/synthetic.json b/internal/providers/configs/synthetic.json index 8e6dbeb9fa2b2f496bfe2deca524497291333bfa..b8986320f659e1f2eb4008eac208391cd1b55268 100644 --- a/internal/providers/configs/synthetic.json +++ b/internal/providers/configs/synthetic.json @@ -55,24 +55,6 @@ "default_reasoning_effort": "medium", "supports_attachments": false }, - { - "id": "hf:zai-org/GLM-4.7", - "name": "GLM 4.7", - "cost_per_1m_in": 0.45, - "cost_per_1m_out": 2.19, - "cost_per_1m_in_cached": 0.45, - "cost_per_1m_out_cached": 0.45, - "context_window": 202752, - "default_max_tokens": 20275, - "can_reason": true, - "reasoning_levels": [ - "low", - "medium", - "high" - ], - "default_reasoning_effort": "medium", - "supports_attachments": false - }, { "id": "hf:zai-org/GLM-4.7-Flash", "name": "GLM 4.7 Flash", diff --git a/internal/providers/configs/venice.json b/internal/providers/configs/venice.json index 70400d1f8941238eb67a7eee23446d64a50765eb..c6db5f474d32359b1728a503e37b8b3933847a84 100644 --- a/internal/providers/configs/venice.json +++ b/internal/providers/configs/venice.json @@ -404,7 +404,7 @@ "supports_attachments": true }, { - "id": "google.gemma-4-26b-a4b-it", + "id": "google-gemma-4-26b-a4b-it", "name": "Google Gemma 4 26B A4B Instruct", "cost_per_1m_in": 0.1625, "cost_per_1m_out": 0.5, @@ -416,7 +416,7 @@ "supports_attachments": true }, { - "id": "google.gemma-4-31b-it", + "id": "google-gemma-4-31b-it", "name": "Google Gemma 4 31B Instruct", "cost_per_1m_in": 0.175, "cost_per_1m_out": 0.5, @@ -440,8 +440,8 @@ "supports_attachments": true }, { - "id": "grok-4-20-beta", - "name": "Grok 4.20 Beta", + "id": "grok-4-20", + "name": "Grok 4.20", "cost_per_1m_in": 2.27, "cost_per_1m_out": 6.8, "cost_per_1m_in_cached": 0,