diff --git a/internal/providers/configs/huggingface.json b/internal/providers/configs/huggingface.json index ed062a6ca3e3816febda46c9d9d94913a7ff6215..2622bcf3b64402a3c03bc69b0913a0bd44318903 100644 --- a/internal/providers/configs/huggingface.json +++ b/internal/providers/configs/huggingface.json @@ -31,6 +31,18 @@ "can_reason": false, "supports_attachments": false }, + { + "id": "Qwen/Qwen3-235B-A22B-Instruct-2507:cerebras", + "name": "Qwen/Qwen3-235B-A22B-Instruct-2507 (cerebras)", + "cost_per_1m_in": 0.6, + "cost_per_1m_out": 1.2, + "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": "Qwen/Qwen3-32B:groq", "name": "Qwen/Qwen3-32B (groq)", diff --git a/internal/providers/configs/openrouter.json b/internal/providers/configs/openrouter.json index b1c9004f0b2d065e920b88f0316962bf714e07c3..7ecd5a664c7de587afee13038361effb4f5a4e7f 100644 --- a/internal/providers/configs/openrouter.json +++ b/internal/providers/configs/openrouter.json @@ -538,10 +538,10 @@ { "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": 0.25, + "cost_per_1m_out": 1, "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, @@ -550,10 +550,10 @@ { "id": "deepseek/deepseek-chat-v3.1", "name": "DeepSeek: DeepSeek V3.1", - "cost_per_1m_in": 0.56, - "cost_per_1m_out": 1.68, + "cost_per_1m_in": 0.6, + "cost_per_1m_out": 1.7, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.28, + "cost_per_1m_out_cached": 0, "context_window": 163840, "default_max_tokens": 16384, "can_reason": true, @@ -586,12 +586,12 @@ { "id": "deepseek/deepseek-v3.2", "name": "DeepSeek: DeepSeek V3.2", - "cost_per_1m_in": 0.269, - "cost_per_1m_out": 0.4, + "cost_per_1m_in": 0.26, + "cost_per_1m_out": 0.38, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.1345, + "cost_per_1m_out_cached": 0.13, "context_window": 163840, - "default_max_tokens": 32768, + "default_max_tokens": 8192, "can_reason": true, "reasoning_levels": [ "low", @@ -607,9 +607,9 @@ "cost_per_1m_in": 0.27, "cost_per_1m_out": 0.41, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.27, + "cost_per_1m_out_cached": 0, "context_window": 163840, - "default_max_tokens": 81920, + "default_max_tokens": 32768, "can_reason": true, "reasoning_levels": [ "low", @@ -640,12 +640,12 @@ { "id": "deepseek/deepseek-v4-pro", "name": "DeepSeek: DeepSeek V4 Pro", - "cost_per_1m_in": 1.7, - "cost_per_1m_out": 3.4, + "cost_per_1m_in": 0.435, + "cost_per_1m_out": 0.87, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.14, + "cost_per_1m_out_cached": 0.00363, "context_window": 1048576, - "default_max_tokens": 32768, + "default_max_tokens": 192000, "can_reason": true, "reasoning_levels": [ "low", @@ -1040,8 +1040,8 @@ "cost_per_1m_out": 1.2, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0.06, - "context_window": 256000, - "default_max_tokens": 40000, + "context_window": 262144, + "default_max_tokens": 72000, "can_reason": false, "supports_attachments": false }, @@ -1060,12 +1060,12 @@ { "id": "meta-llama/llama-3.3-70b-instruct", "name": "Meta: Llama 3.3 70B Instruct", - "cost_per_1m_in": 0.6, - "cost_per_1m_out": 0.6, + "cost_per_1m_in": 0.13, + "cost_per_1m_out": 0.4, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.6, + "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 65536, + "default_max_tokens": 13107, "can_reason": false, "supports_attachments": false }, @@ -1093,33 +1093,15 @@ "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": 1.02, + "cost_per_1m_out": 1, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, - "context_window": 204800, - "default_max_tokens": 65536, + "cost_per_1m_out_cached": 0.03, + "context_window": 196608, + "default_max_tokens": 98304, "can_reason": true, "reasoning_levels": [ "low", @@ -1155,7 +1137,7 @@ "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", @@ -1288,12 +1270,12 @@ { "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 }, @@ -1474,12 +1456,12 @@ { "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": 0.6, + "cost_per_1m_out_cached": 0.5, "context_window": 262144, - "default_max_tokens": 4096, + "default_max_tokens": 8192, "can_reason": false, "supports_attachments": false }, @@ -1489,7 +1471,7 @@ "cost_per_1m_in": 0.6, "cost_per_1m_out": 2.5, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.6, "context_window": 262144, "default_max_tokens": 131072, "can_reason": true, @@ -1504,10 +1486,10 @@ { "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": 0.44, + "cost_per_1m_out": 2.2, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.07, + "cost_per_1m_out_cached": 0.22, "context_window": 262144, "default_max_tokens": 131072, "can_reason": true, @@ -1522,12 +1504,12 @@ { "id": "moonshotai/kimi-k2.6", "name": "MoonshotAI: Kimi K2.6", - "cost_per_1m_in": 0.9, + "cost_per_1m_in": 0.95, "cost_per_1m_out": 4, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.2, + "cost_per_1m_out_cached": 0.16, "context_window": 262144, - "default_max_tokens": 131072, + "default_max_tokens": 26214, "can_reason": true, "reasoning_levels": [ "low", @@ -1825,7 +1807,7 @@ "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, @@ -1963,9 +1945,9 @@ "cost_per_1m_in": 0.05, "cost_per_1m_out": 0.4, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.01, + "cost_per_1m_out_cached": 0.005, "context_window": 400000, - "default_max_tokens": 40000, + "default_max_tokens": 64000, "can_reason": true, "reasoning_levels": [ "low", @@ -2284,10 +2266,10 @@ { "id": "openai/gpt-oss-120b", "name": "OpenAI: gpt-oss-120b", - "cost_per_1m_in": 0.15, - "cost_per_1m_out": 0.6, + "cost_per_1m_in": 0.1, + "cost_per_1m_out": 0.49, "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, "can_reason": true, @@ -2704,12 +2686,12 @@ { "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": 8192, + "context_window": 131072, + "default_max_tokens": 4096, "can_reason": true, "reasoning_levels": [ "low", @@ -2791,20 +2773,20 @@ "cost_per_1m_in": 0.1, "cost_per_1m_out": 0.3, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.1, + "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 131072, + "default_max_tokens": 26214, "can_reason": false, "supports_attachments": false }, { "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": [ @@ -2855,11 +2837,11 @@ "id": "qwen/qwen3-coder-30b-a3b-instruct", "name": "Qwen: Qwen3 Coder 30B A3B Instruct", "cost_per_1m_in": 0.07, - "cost_per_1m_out": 0.27, + "cost_per_1m_out": 0.28, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, - "context_window": 160000, - "default_max_tokens": 16384, + "context_window": 262144, + "default_max_tokens": 131072, "can_reason": false, "supports_attachments": false }, @@ -2956,12 +2938,12 @@ { "id": "qwen/qwen3-next-80b-a3b-instruct", "name": "Qwen: Qwen3 Next 80B A3B Instruct", - "cost_per_1m_in": 0.09, + "cost_per_1m_in": 0.1, "cost_per_1m_out": 1.1, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.07, "context_window": 262144, - "default_max_tokens": 8192, + "default_max_tokens": 131072, "can_reason": false, "supports_attachments": false }, @@ -2980,12 +2962,12 @@ { "id": "qwen/qwen3-next-80b-a3b-thinking", "name": "Qwen: Qwen3 Next 80B A3B Thinking", - "cost_per_1m_in": 0.15, - "cost_per_1m_out": 1.2, + "cost_per_1m_in": 0.0975, + "cost_per_1m_out": 0.78, "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", @@ -3103,9 +3085,9 @@ "cost_per_1m_in": 0.6, "cost_per_1m_out": 3.6, "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": 131072, + "default_max_tokens": 32768, "can_reason": true, "reasoning_levels": [ "low", @@ -3154,8 +3136,8 @@ { "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": 0.4, + "cost_per_1m_out": 3.2, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 262144, @@ -3172,10 +3154,10 @@ { "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, @@ -3190,12 +3172,12 @@ { "id": "qwen/qwen3.5-35b-a3b", "name": "Qwen: Qwen3.5-35B-A3B", - "cost_per_1m_in": 0.2, - "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": 0.05, + "cost_per_1m_out_cached": 0.225, "context_window": 262144, - "default_max_tokens": 131072, + "default_max_tokens": 32768, "can_reason": true, "reasoning_levels": [ "low", @@ -3244,12 +3226,12 @@ { "id": "qwen/qwen3.6-27b", "name": "Qwen: Qwen3.6 27B", - "cost_per_1m_in": 0.6, - "cost_per_1m_out": 3.6, + "cost_per_1m_in": 0.32, + "cost_per_1m_out": 3.2, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 32768, + "default_max_tokens": 40960, "can_reason": true, "reasoning_levels": [ "low", @@ -3333,7 +3315,7 @@ "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0.02, "context_window": 262144, - "default_max_tokens": 131072, + "default_max_tokens": 8192, "can_reason": true, "reasoning_levels": [ "low", @@ -3445,9 +3427,9 @@ "cost_per_1m_in": 0.1, "cost_per_1m_out": 0.3, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.01, + "cost_per_1m_out_cached": 0.02, "context_window": 262144, - "default_max_tokens": 32768, + "default_max_tokens": 16000, "can_reason": true, "reasoning_levels": [ "low", @@ -3549,7 +3531,7 @@ "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0.11, "context_window": 131072, - "default_max_tokens": 49152, + "default_max_tokens": 48000, "can_reason": true, "reasoning_levels": [ "low", @@ -3616,12 +3598,12 @@ { "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", @@ -3670,12 +3652,12 @@ { "id": "z-ai/glm-4.7-flash", "name": "Z.ai: GLM 4.7 Flash", - "cost_per_1m_in": 0.1, - "cost_per_1m_out": 0.43, + "cost_per_1m_in": 0.06, + "cost_per_1m_out": 0.4, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.01, "context_window": 202752, - "default_max_tokens": 101376, + "default_max_tokens": 8192, "can_reason": true, "reasoning_levels": [ "low", @@ -3784,10 +3766,10 @@ { "id": "x-ai/grok-3", "name": "xAI: Grok 3", - "cost_per_1m_in": 3, - "cost_per_1m_out": 15, + "cost_per_1m_in": 5, + "cost_per_1m_out": 25, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.75, + "cost_per_1m_out_cached": 1.25, "context_window": 131072, "default_max_tokens": 13107, "can_reason": false, @@ -3796,10 +3778,10 @@ { "id": "x-ai/grok-3-beta", "name": "xAI: Grok 3 Beta", - "cost_per_1m_in": 3, - "cost_per_1m_out": 15, + "cost_per_1m_in": 5, + "cost_per_1m_out": 25, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.75, + "cost_per_1m_out_cached": 1.25, "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 0d6a310660d8a655de908c46ff25890a8895b519..e6650198b6cc5fc79bed68b6a0bb51c97c4f9f03 100644 --- a/internal/providers/configs/synthetic.json +++ b/internal/providers/configs/synthetic.json @@ -127,6 +127,18 @@ "default_reasoning_effort": "medium", "supports_attachments": false }, + { + "id": "hf:moonshotai/Kimi-K2.5", + "name": "Kimi K2.5", + "cost_per_1m_in": 0.5, + "cost_per_1m_out": 2.8, + "cost_per_1m_in_cached": 0.5, + "cost_per_1m_out_cached": 0.5, + "context_window": 262144, + "default_max_tokens": 26214, + "can_reason": false, + "supports_attachments": true + }, { "id": "hf:moonshotai/Kimi-K2.6", "name": "Kimi K2.6",