diff --git a/internal/providers/configs/huggingface.json b/internal/providers/configs/huggingface.json index 9d4189a85dc7fb887bc7e572944703b65805a999..1caa40210683531cfe294825642f05ec6144c698 100644 --- a/internal/providers/configs/huggingface.json +++ b/internal/providers/configs/huggingface.json @@ -111,19 +111,6 @@ "supports_attachments": false, "options": {} }, - { - "id": "deepseek-ai/DeepSeek-V3.2:fireworks-ai", - "name": "deepseek-ai/DeepSeek-V3.2 (fireworks-ai)", - "cost_per_1m_in": 0, - "cost_per_1m_out": 0, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, - "context_window": 163840, - "default_max_tokens": 8192, - "can_reason": false, - "supports_attachments": false, - "options": {} - }, { "id": "meta-llama/Llama-3.3-70B-Instruct:groq", "name": "meta-llama/Llama-3.3-70B-Instruct (groq)", diff --git a/internal/providers/configs/openrouter.json b/internal/providers/configs/openrouter.json index 002450502d367bf73d4e5e139d5fda46eb88925a..e54224c91c4d8beb4961dba448edfb0bb0819b69 100644 --- a/internal/providers/configs/openrouter.json +++ b/internal/providers/configs/openrouter.json @@ -547,29 +547,23 @@ { "id": "deepseek/deepseek-chat-v3-0324", "name": "DeepSeek: DeepSeek V3 0324", - "cost_per_1m_in": 0.77, - "cost_per_1m_out": 0.77, + "cost_per_1m_in": 0.27, + "cost_per_1m_out": 1.12, "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": 65536, - "can_reason": true, - "reasoning_levels": [ - "low", - "medium", - "high" - ], - "default_reasoning_effort": "medium", + "default_max_tokens": 81920, + "can_reason": false, "supports_attachments": false, "options": {} }, { "id": "deepseek/deepseek-chat-v3.1", "name": "DeepSeek: DeepSeek V3.1", - "cost_per_1m_in": 0.21, - "cost_per_1m_out": 0.79, + "cost_per_1m_in": 0.56, + "cost_per_1m_out": 1.68, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.13, + "cost_per_1m_out_cached": 0.28, "context_window": 163840, "default_max_tokens": 16384, "can_reason": true, @@ -661,12 +655,12 @@ { "id": "deepseek/deepseek-r1-0528", "name": "DeepSeek: R1 0528", - "cost_per_1m_in": 0.7, - "cost_per_1m_out": 2.5, + "cost_per_1m_in": 0.45, + "cost_per_1m_out": 2.15, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.35, + "cost_per_1m_out_cached": 0.225, "context_window": 163840, - "default_max_tokens": 16384, + "default_max_tokens": 32768, "can_reason": true, "reasoning_levels": [ "low", @@ -1012,12 +1006,12 @@ { "id": "meta-llama/llama-3.3-70b-instruct", "name": "Meta: Llama 3.3 70B Instruct", - "cost_per_1m_in": 0.88, - "cost_per_1m_out": 0.88, + "cost_per_1m_in": 0.13, + "cost_per_1m_out": 0.4, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 1024, + "default_max_tokens": 64000, "can_reason": false, "supports_attachments": false, "options": {} @@ -1253,9 +1247,9 @@ "cost_per_1m_in": 0.2, "cost_per_1m_out": 0.2, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.02, + "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 26214, + "default_max_tokens": 131072, "can_reason": false, "supports_attachments": true, "options": {} @@ -1490,12 +1484,12 @@ { "id": "moonshotai/kimi-k2-0905", "name": "MoonshotAI: Kimi K2 0905", - "cost_per_1m_in": 0.6, - "cost_per_1m_out": 2.5, + "cost_per_1m_in": 1, + "cost_per_1m_out": 3, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.3, + "cost_per_1m_out_cached": 0.5, "context_window": 262144, - "default_max_tokens": 26214, + "default_max_tokens": 8192, "can_reason": false, "supports_attachments": false, "options": {} @@ -1503,8 +1497,8 @@ { "id": "moonshotai/kimi-k2-thinking", "name": "MoonshotAI: Kimi K2 Thinking", - "cost_per_1m_in": 1.15, - "cost_per_1m_out": 8, + "cost_per_1m_in": 0.6, + "cost_per_1m_out": 2.5, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0.15, "context_window": 262144, @@ -1522,12 +1516,12 @@ { "id": "moonshotai/kimi-k2.5", "name": "MoonshotAI: Kimi K2.5", - "cost_per_1m_in": 0.45, - "cost_per_1m_out": 2.25, + "cost_per_1m_in": 0.6, + "cost_per_1m_out": 3, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.07, + "cost_per_1m_out_cached": 0.1, "context_window": 262144, - "default_max_tokens": 131072, + "default_max_tokens": 26214, "can_reason": true, "reasoning_levels": [ "low", @@ -2037,7 +2031,7 @@ "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, @@ -2302,11 +2296,11 @@ "id": "openai/gpt-oss-120b", "name": "OpenAI: gpt-oss-120b", "cost_per_1m_in": 0.05, - "cost_per_1m_out": 0.25, + "cost_per_1m_out": 0.45, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.025, "context_window": 131072, - "default_max_tokens": 16384, + "default_max_tokens": 32768, "can_reason": true, "reasoning_levels": [ "low", @@ -2339,12 +2333,12 @@ { "id": "openai/gpt-oss-20b", "name": "OpenAI: gpt-oss-20b", - "cost_per_1m_in": 0.03, - "cost_per_1m_out": 0.14, + "cost_per_1m_in": 0.045, + "cost_per_1m_out": 0.18, "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", @@ -2906,12 +2900,12 @@ { "id": "qwen/qwen3-coder", "name": "Qwen: Qwen3 Coder 480B A35B", - "cost_per_1m_in": 0.22, - "cost_per_1m_out": 1, + "cost_per_1m_in": 0.3, + "cost_per_1m_out": 1.3, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.022, + "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 26214, + "default_max_tokens": 32768, "can_reason": false, "supports_attachments": false, "options": {} @@ -3003,12 +2997,12 @@ { "id": "qwen/qwen3-next-80b-a3b-instruct", "name": "Qwen: Qwen3 Next 80B A3B Instruct", - "cost_per_1m_in": 0.15, - "cost_per_1m_out": 1.5, + "cost_per_1m_in": 0.1, + "cost_per_1m_out": 1.1, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.15, + "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 65536, + "default_max_tokens": 131072, "can_reason": false, "supports_attachments": false, "options": {} @@ -3061,12 +3055,12 @@ { "id": "qwen/qwen3-vl-235b-a22b-thinking", "name": "Qwen: Qwen3 VL 235B A22B Thinking", - "cost_per_1m_in": 0.45, - "cost_per_1m_out": 3.5, + "cost_per_1m_in": 0.26, + "cost_per_1m_out": 2.6, "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": true, "reasoning_levels": [ "low", @@ -3093,12 +3087,12 @@ { "id": "qwen/qwen3-vl-30b-a3b-thinking", "name": "Qwen: Qwen3 VL 30B A3B Thinking", - "cost_per_1m_in": 0.29, + "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": 262144, - "default_max_tokens": 131072, + "context_window": 131072, + "default_max_tokens": 16384, "can_reason": true, "reasoning_levels": [ "low", @@ -3195,10 +3189,10 @@ { "id": "qwen/qwen3.5-122b-a10b", "name": "Qwen: Qwen3.5-122B-A10B", - "cost_per_1m_in": 0.4, - "cost_per_1m_out": 3.2, + "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, @@ -3214,10 +3208,10 @@ { "id": "qwen/qwen3.5-27b", "name": "Qwen: Qwen3.5-27B", - "cost_per_1m_in": 0.3, - "cost_per_1m_out": 2.4, + "cost_per_1m_in": 0.27, + "cost_per_1m_out": 2.16, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.27, "context_window": 262144, "default_max_tokens": 32768, "can_reason": true, @@ -3424,12 +3418,12 @@ { "id": "xiaomi/mimo-v2-flash", "name": "Xiaomi: MiMo-V2-Flash", - "cost_per_1m_in": 0.09, - "cost_per_1m_out": 0.29, + "cost_per_1m_in": 0.1, + "cost_per_1m_out": 0.3, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.045, + "cost_per_1m_out_cached": 0.1, "context_window": 262144, - "default_max_tokens": 32768, + "default_max_tokens": 131072, "can_reason": true, "reasoning_levels": [ "low", @@ -3570,12 +3564,12 @@ { "id": "z-ai/glm-4.6", "name": "Z.ai: GLM 4.6", - "cost_per_1m_in": 0.39, - "cost_per_1m_out": 1.9, + "cost_per_1m_in": 0.55, + "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": 204800, - "default_max_tokens": 102400, + "default_max_tokens": 65536, "can_reason": true, "reasoning_levels": [ "low", @@ -3684,10 +3678,10 @@ { "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, @@ -3697,10 +3691,10 @@ { "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, @@ -3710,10 +3704,10 @@ { "id": "x-ai/grok-3-mini", "name": "xAI: Grok 3 Mini", - "cost_per_1m_in": 0.6, - "cost_per_1m_out": 4, + "cost_per_1m_in": 0.3, + "cost_per_1m_out": 0.5, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.15, + "cost_per_1m_out_cached": 0.075, "context_window": 131072, "default_max_tokens": 13107, "can_reason": true, @@ -3729,10 +3723,10 @@ { "id": "x-ai/grok-3-mini-beta", "name": "xAI: Grok 3 Mini Beta", - "cost_per_1m_in": 0.6, - "cost_per_1m_out": 4, + "cost_per_1m_in": 0.3, + "cost_per_1m_out": 0.5, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.15, + "cost_per_1m_out_cached": 0.075, "context_window": 131072, "default_max_tokens": 13107, "can_reason": true,