diff --git a/internal/providers/configs/huggingface.json b/internal/providers/configs/huggingface.json index 9d4189a85dc7fb887bc7e572944703b65805a999..f4c0c1707fbd50ffef858a87cf274cfbac3df6ee 100644 --- a/internal/providers/configs/huggingface.json +++ b/internal/providers/configs/huggingface.json @@ -20,19 +20,6 @@ "supports_attachments": false, "options": {} }, - { - "id": "MiniMaxAI/MiniMax-M2.5:fireworks-ai", - "name": "MiniMaxAI/MiniMax-M2.5 (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": 204800, - "default_max_tokens": 8192, - "can_reason": false, - "supports_attachments": false, - "options": {} - }, { "id": "Qwen/Qwen3-32B:groq", "name": "Qwen/Qwen3-32B (groq)", diff --git a/internal/providers/configs/openrouter.json b/internal/providers/configs/openrouter.json index a175c1e8e73183dec4e1eacf51e7e8e2a22f077d..feb06a6c4deac693eb6ddeee8070504adf50b25a 100644 --- a/internal/providers/configs/openrouter.json +++ b/internal/providers/configs/openrouter.json @@ -547,23 +547,29 @@ { "id": "deepseek/deepseek-chat-v3-0324", "name": "DeepSeek: DeepSeek V3 0324", - "cost_per_1m_in": 0.27, - "cost_per_1m_out": 1.12, + "cost_per_1m_in": 0.77, + "cost_per_1m_out": 0.77, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.135, + "cost_per_1m_out_cached": 0, "context_window": 163840, - "default_max_tokens": 81920, - "can_reason": false, + "default_max_tokens": 65536, + "can_reason": true, + "reasoning_levels": [ + "low", + "medium", + "high" + ], + "default_reasoning_effort": "medium", "supports_attachments": false, "options": {} }, { "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": 0.21, + "cost_per_1m_out": 0.79, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.13, "context_window": 163840, "default_max_tokens": 16384, "can_reason": true, @@ -832,7 +838,7 @@ "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", @@ -998,7 +1004,7 @@ "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, "options": {} @@ -1006,12 +1012,12 @@ { "id": "meta-llama/llama-3.3-70b-instruct", "name": "Meta: Llama 3.3 70B Instruct", - "cost_per_1m_in": 0.135, - "cost_per_1m_out": 0.4, + "cost_per_1m_in": 0.88, + "cost_per_1m_out": 0.88, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 60000, + "default_max_tokens": 1024, "can_reason": false, "supports_attachments": false, "options": {} @@ -1080,9 +1086,9 @@ "cost_per_1m_in": 0.3, "cost_per_1m_out": 1.2, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.03, - "context_window": 204800, - "default_max_tokens": 65536, + "cost_per_1m_out_cached": 0.15, + "context_window": 196608, + "default_max_tokens": 98304, "can_reason": true, "reasoning_levels": [ "low", @@ -1096,12 +1102,12 @@ { "id": "minimax/minimax-m2.5", "name": "MiniMax: MiniMax M2.5", - "cost_per_1m_in": 0.6, - "cost_per_1m_out": 2.4, + "cost_per_1m_in": 0.34, + "cost_per_1m_out": 1.19, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.06, - "context_window": 204800, - "default_max_tokens": 65536, + "cost_per_1m_out_cached": 0.04, + "context_window": 198000, + "default_max_tokens": 16384, "can_reason": true, "reasoning_levels": [ "low", @@ -1244,12 +1250,12 @@ { "id": "mistralai/ministral-14b-2512", "name": "Mistral: Ministral 3 14B 2512", - "cost_per_1m_in": 0.2, - "cost_per_1m_out": 0.2, + "cost_per_1m_in": 0.3, + "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, "context_window": 262144, - "default_max_tokens": 26214, + "default_max_tokens": 131072, "can_reason": false, "supports_attachments": true, "options": {} @@ -1257,12 +1263,12 @@ { "id": "mistralai/ministral-3b-2512", "name": "Mistral: Ministral 3 3B 2512", - "cost_per_1m_in": 0.1, - "cost_per_1m_out": 0.1, + "cost_per_1m_in": 0.35, + "cost_per_1m_out": 0.35, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.01, + "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 13107, + "default_max_tokens": 65536, "can_reason": false, "supports_attachments": true, "options": {} @@ -1502,7 +1508,7 @@ "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 131072, + "default_max_tokens": 26214, "can_reason": true, "reasoning_levels": [ "low", @@ -1517,11 +1523,11 @@ "id": "moonshotai/kimi-k2.5", "name": "MoonshotAI: Kimi K2.5", "cost_per_1m_in": 0.6, - "cost_per_1m_out": 2.8, + "cost_per_1m_out": 3, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.1, "context_window": 262144, - "default_max_tokens": 131072, + "default_max_tokens": 26214, "can_reason": true, "reasoning_levels": [ "low", @@ -1700,8 +1706,8 @@ { "id": "nex-agi/deepseek-v3.1-nex-n1", "name": "Nex AGI: DeepSeek V3.1 Nex N1", - "cost_per_1m_in": 0.27, - "cost_per_1m_out": 1, + "cost_per_1m_in": 0.135, + "cost_per_1m_out": 0.5, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, @@ -1770,7 +1776,7 @@ "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0.1, "context_window": 1047576, - "default_max_tokens": 104757, + "default_max_tokens": 16384, "can_reason": false, "supports_attachments": true, "options": {} @@ -1794,7 +1800,7 @@ "cost_per_1m_in": 2.5, "cost_per_1m_out": 10, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 1.25, "context_window": 128000, "default_max_tokens": 8192, "can_reason": false, @@ -2031,7 +2037,7 @@ "cost_per_1m_in": 1.25, "cost_per_1m_out": 10, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.13, + "cost_per_1m_out_cached": 0.125, "context_window": 400000, "default_max_tokens": 64000, "can_reason": true, @@ -2141,7 +2147,7 @@ "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0.175, "context_window": 128000, - "default_max_tokens": 16000, + "default_max_tokens": 8192, "can_reason": false, "supports_attachments": true, "options": {} @@ -2296,11 +2302,11 @@ "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", @@ -2333,12 +2339,12 @@ { "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", @@ -2709,12 +2715,12 @@ { "id": "qwen/qwen3-14b", "name": "Qwen: Qwen3 14B", - "cost_per_1m_in": 0.12, - "cost_per_1m_out": 0.24, + "cost_per_1m_in": 0.2275, + "cost_per_1m_out": 0.91, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, - "context_window": 40960, - "default_max_tokens": 20480, + "context_window": 131072, + "default_max_tokens": 4096, "can_reason": true, "reasoning_levels": [ "low", @@ -2900,8 +2906,8 @@ { "id": "qwen/qwen3-coder", "name": "Qwen: Qwen3 Coder 480B A35B", - "cost_per_1m_in": 0.22, - "cost_per_1m_out": 1.8, + "cost_per_1m_in": 0.3, + "cost_per_1m_out": 1.3, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 262144, @@ -2939,12 +2945,12 @@ { "id": "qwen/qwen3-coder-next", "name": "Qwen: Qwen3 Coder Next", - "cost_per_1m_in": 0.12, - "cost_per_1m_out": 0.75, + "cost_per_1m_in": 0.15, + "cost_per_1m_out": 0.8, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.06, + "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 32768, + "default_max_tokens": 131072, "can_reason": false, "supports_attachments": false, "options": {} @@ -2997,8 +3003,8 @@ { "id": "qwen/qwen3-next-80b-a3b-instruct", "name": "Qwen: Qwen3 Next 80B A3B Instruct", - "cost_per_1m_in": 0.1, - "cost_per_1m_out": 1.1, + "cost_per_1m_in": 0.15, + "cost_per_1m_out": 1.2, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 262144, @@ -3023,12 +3029,12 @@ { "id": "qwen/qwen3-next-80b-a3b-thinking", "name": "Qwen: Qwen3 Next 80B A3B Thinking", - "cost_per_1m_in": 0.3, - "cost_per_1m_out": 0.3, + "cost_per_1m_in": 0.15, + "cost_per_1m_out": 1.5, "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", @@ -3042,8 +3048,8 @@ { "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": 0.26, + "cost_per_1m_out": 1.04, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, @@ -3189,10 +3195,10 @@ { "id": "qwen/qwen3.5-122b-a10b", "name": "Qwen: Qwen3.5-122B-A10B", - "cost_per_1m_in": 0.3, - "cost_per_1m_out": 2.4, + "cost_per_1m_in": 0.26, + "cost_per_1m_out": 2.08, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.3, + "cost_per_1m_out_cached": 0, "context_window": 262144, "default_max_tokens": 32768, "can_reason": true, @@ -3208,10 +3214,10 @@ { "id": "qwen/qwen3.5-27b", "name": "Qwen: Qwen3.5-27B", - "cost_per_1m_in": 0.27, - "cost_per_1m_out": 2.16, + "cost_per_1m_in": 0.3, + "cost_per_1m_out": 2.4, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.27, + "cost_per_1m_out_cached": 0, "context_window": 262144, "default_max_tokens": 32768, "can_reason": true, @@ -3227,12 +3233,12 @@ { "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": 0.225, + "cost_per_1m_out": 1.8, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.225, "context_window": 262144, - "default_max_tokens": 131072, + "default_max_tokens": 32768, "can_reason": true, "reasoning_levels": [ "low", @@ -3507,12 +3513,12 @@ { "id": "z-ai/glm-4.5-air", "name": "Z.ai: GLM 4.5 Air", - "cost_per_1m_in": 0.2, - "cost_per_1m_out": 1.2, + "cost_per_1m_in": 0.13, + "cost_per_1m_out": 0.85, "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": 13107, + "default_max_tokens": 49152, "can_reason": true, "reasoning_levels": [ "low", @@ -3678,10 +3684,10 @@ { "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, @@ -3691,10 +3697,10 @@ { "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,