diff --git a/internal/providers/configs/huggingface.json b/internal/providers/configs/huggingface.json index b95b64bb866f776102fe3aba6462822a50d5ad75..6256bab6cbce3d48382a378dd38a945eb10108e9 100644 --- a/internal/providers/configs/huggingface.json +++ b/internal/providers/configs/huggingface.json @@ -7,32 +7,6 @@ "default_large_model_id": "moonshotai/Kimi-K2-Instruct-0905:groq", "default_small_model_id": "openai/gpt-oss-20b:groq", "models": [ - { - "id": "Qwen/Qwen3-235B-A22B-Instruct-2507:fireworks-ai", - "name": "Qwen/Qwen3-235B-A22B-Instruct-2507 (fireworks-ai)", - "cost_per_1m_in": 0.22, - "cost_per_1m_out": 0.88, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, - "context_window": 262144, - "default_max_tokens": 8192, - "can_reason": false, - "supports_attachments": false, - "options": {} - }, - { - "id": "Qwen/Qwen3-235B-A22B-Thinking-2507:fireworks-ai", - "name": "Qwen/Qwen3-235B-A22B-Thinking-2507 (fireworks-ai)", - "cost_per_1m_in": 0.22, - "cost_per_1m_out": 0.88, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, - "context_window": 262144, - "default_max_tokens": 8192, - "can_reason": false, - "supports_attachments": false, - "options": {} - }, { "id": "Qwen/Qwen3-32B:groq", "name": "Qwen/Qwen3-32B (groq)", @@ -46,19 +20,6 @@ "supports_attachments": false, "options": {} }, - { - "id": "Qwen/Qwen3-Coder-480B-A35B-Instruct:cerebras", - "name": "Qwen/Qwen3-Coder-480B-A35B-Instruct (cerebras)", - "cost_per_1m_in": 2, - "cost_per_1m_out": 2, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, - "context_window": 262144, - "default_max_tokens": 8192, - "can_reason": false, - "supports_attachments": false, - "options": {} - }, { "id": "deepcogito/cogito-671b-v2.1:fireworks-ai", "name": "deepcogito/cogito-671b-v2.1 (fireworks-ai)", @@ -227,32 +188,6 @@ "can_reason": false, "supports_attachments": false, "options": {} - }, - { - "id": "zai-org/GLM-4.6:cerebras", - "name": "zai-org/GLM-4.6 (cerebras)", - "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": "zai-org/GLM-5:fireworks-ai", - "name": "zai-org/GLM-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": 202752, - "default_max_tokens": 8192, - "can_reason": false, - "supports_attachments": false, - "options": {} } ], "default_headers": { diff --git a/internal/providers/configs/openrouter.json b/internal/providers/configs/openrouter.json index f75b59175babca59be577d0e16a3867fae7036ed..7a49ede0889e6e3afa0e87e38e5e819bf73257d3 100644 --- a/internal/providers/configs/openrouter.json +++ b/internal/providers/configs/openrouter.json @@ -135,8 +135,8 @@ "name": "Anthropic: Claude 3.5 Sonnet", "cost_per_1m_in": 6, "cost_per_1m_out": 30, - "cost_per_1m_in_cached": 7.5, - "cost_per_1m_out_cached": 0.6, + "cost_per_1m_in_cached": 0, + "cost_per_1m_out_cached": 0, "context_window": 200000, "default_max_tokens": 4096, "can_reason": false, @@ -528,25 +528,31 @@ { "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": 0.19999999999999998, - "cost_per_1m_out": 0.7999999999999999, + "cost_per_1m_in": 0.21, + "cost_per_1m_out": 0.7899999999999999, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.09999999999999999, + "cost_per_1m_out_cached": 0.1300000002, "context_window": 163840, - "default_max_tokens": 32768, + "default_max_tokens": 16384, "can_reason": true, "reasoning_levels": [ "low", @@ -819,7 +825,7 @@ "cost_per_1m_in_cached": 0.08333333333333334, "cost_per_1m_out_cached": 0.049999999999999996, "context_window": 1048576, - "default_max_tokens": 32768, + "default_max_tokens": 32767, "can_reason": true, "reasoning_levels": [ "low", @@ -887,6 +893,19 @@ "supports_attachments": true, "options": {} }, + { + "id": "google/gemma-3-27b-it:free", + "name": "Google: Gemma 3 27B (free)", + "cost_per_1m_in": 0, + "cost_per_1m_out": 0, + "cost_per_1m_in_cached": 0, + "cost_per_1m_out_cached": 0, + "context_window": 128000, + "default_max_tokens": 64000, + "can_reason": false, + "supports_attachments": true, + "options": {} + }, { "id": "inception/mercury", "name": "Inception: Mercury", @@ -960,7 +979,7 @@ "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 8192, + "default_max_tokens": 13107, "can_reason": false, "supports_attachments": false, "options": {} @@ -968,12 +987,12 @@ { "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": 0.39999999999999997, + "cost_per_1m_in": 0.25, + "cost_per_1m_out": 0.75, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 60000, + "default_max_tokens": 13107, "can_reason": false, "supports_attachments": false, "options": {} @@ -1007,16 +1026,35 @@ { "id": "meta-llama/llama-4-scout", "name": "Meta: Llama 4 Scout", - "cost_per_1m_in": 0.25, - "cost_per_1m_out": 0.7, + "cost_per_1m_in": 0.11, + "cost_per_1m_out": 0.33999999999999997, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, - "context_window": 1310720, + "context_window": 131072, "default_max_tokens": 4096, "can_reason": false, "supports_attachments": true, "options": {} }, + { + "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, + "options": {} + }, { "id": "minimax/minimax-m2", "name": "MiniMax: MiniMax M2", @@ -1040,7 +1078,7 @@ "id": "minimax/minimax-m2.1", "name": "MiniMax: MiniMax M2.1", "cost_per_1m_in": 0.3, - "cost_per_1m_out": 1.2, + "cost_per_1m_out": 2.4, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0.03, "context_window": 204800, @@ -1058,12 +1096,12 @@ { "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": 0.3, + "cost_per_1m_out": 1.2, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.06, + "cost_per_1m_out_cached": 0.03, "context_window": 204800, - "default_max_tokens": 65536, + "default_max_tokens": 65550, "can_reason": true, "reasoning_levels": [ "low", @@ -1285,12 +1323,12 @@ { "id": "mistralai/mistral-small-3.2-24b-instruct", "name": "Mistral: Mistral Small 3.2 24B", - "cost_per_1m_in": 0.06, - "cost_per_1m_out": 0.18, + "cost_per_1m_in": 0.09999999999999999, + "cost_per_1m_out": 0.3, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.03, + "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 65536, + "default_max_tokens": 13107, "can_reason": false, "supports_attachments": true, "options": {} @@ -1402,10 +1440,10 @@ { "id": "moonshotai/kimi-k2-0905:exacto", "name": "MoonshotAI: Kimi K2 0905 (exacto)", - "cost_per_1m_in": 0.6, - "cost_per_1m_out": 2.5, + "cost_per_1m_in": 1.15, + "cost_per_1m_out": 8, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.15, "context_window": 262144, "default_max_tokens": 26214, "can_reason": false, @@ -1415,10 +1453,10 @@ { "id": "moonshotai/kimi-k2-thinking", "name": "MoonshotAI: Kimi K2 Thinking", - "cost_per_1m_in": 0.6, + "cost_per_1m_in": 0.55, "cost_per_1m_out": 2.5, "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": 131072, "can_reason": true, @@ -1435,11 +1473,11 @@ "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", @@ -2086,12 +2124,12 @@ { "id": "openai/gpt-oss-120b", "name": "OpenAI: gpt-oss-120b", - "cost_per_1m_in": 0.04, - "cost_per_1m_out": 0.18, + "cost_per_1m_in": 0.15, + "cost_per_1m_out": 0.6, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.02, + "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 32768, + "default_max_tokens": 13107, "can_reason": true, "reasoning_levels": [ "low", @@ -2105,12 +2143,12 @@ { "id": "openai/gpt-oss-120b:exacto", "name": "OpenAI: gpt-oss-120b (exacto)", - "cost_per_1m_in": 0.039, - "cost_per_1m_out": 0.19, + "cost_per_1m_in": 0.15, + "cost_per_1m_out": 0.6, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.075, "context_window": 131072, - "default_max_tokens": 13107, + "default_max_tokens": 32768, "can_reason": true, "reasoning_levels": [ "low", @@ -2382,6 +2420,25 @@ "supports_attachments": false, "options": {} }, + { + "id": "qwen/qwq-32b", + "name": "Qwen: QwQ 32B", + "cost_per_1m_in": 0.15, + "cost_per_1m_out": 0.58, + "cost_per_1m_in_cached": 0, + "cost_per_1m_out_cached": 0, + "context_window": 131072, + "default_max_tokens": 65536, + "can_reason": true, + "reasoning_levels": [ + "low", + "medium", + "high" + ], + "default_reasoning_effort": "medium", + "supports_attachments": false, + "options": {} + }, { "id": "qwen/qwen-plus-2025-07-28", "name": "Qwen: Qwen Plus 0728", @@ -2533,12 +2590,12 @@ { "id": "qwen/qwen3-235b-a22b-thinking-2507", "name": "Qwen: Qwen3 235B A22B Thinking 2507", - "cost_per_1m_in": 0.6, - "cost_per_1m_out": 2.4, + "cost_per_1m_in": 0.65, + "cost_per_1m_out": 3, "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", @@ -2584,12 +2641,12 @@ { "id": "qwen/qwen3-30b-a3b-thinking-2507", "name": "Qwen: Qwen3 30B A3B Thinking 2507", - "cost_per_1m_in": 0.09999999999999999, + "cost_per_1m_in": 0.09, "cost_per_1m_out": 0.3, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 26214, + "default_max_tokens": 65536, "can_reason": true, "reasoning_levels": [ "low", @@ -2660,12 +2717,12 @@ { "id": "qwen/qwen3-coder-30b-a3b-instruct", "name": "Qwen: Qwen3 Coder 30B A3B Instruct", - "cost_per_1m_in": 0.07, - "cost_per_1m_out": 0.28, + "cost_per_1m_in": 0.09999999999999999, + "cost_per_1m_out": 0.3, "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": false, "supports_attachments": false, "options": {} @@ -2725,12 +2782,12 @@ { "id": "qwen/qwen3-coder-next", "name": "Qwen: Qwen3 Coder Next", - "cost_per_1m_in": 0.18, - "cost_per_1m_out": 1.35, + "cost_per_1m_in": 0.19999999999999998, + "cost_per_1m_out": 1.5, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.18, + "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 131072, + "default_max_tokens": 32768, "can_reason": false, "supports_attachments": false, "options": {} @@ -2783,12 +2840,12 @@ { "id": "qwen/qwen3-next-80b-a3b-instruct", "name": "Qwen: Qwen3 Next 80B A3B Instruct", - "cost_per_1m_in": 0.09999999999999999, + "cost_per_1m_in": 0.09, "cost_per_1m_out": 1.1, "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": false, "supports_attachments": false, "options": {} @@ -2828,8 +2885,8 @@ { "id": "qwen/qwen3-vl-235b-a22b-instruct", "name": "Qwen: Qwen3 VL 235B A22B Instruct", - "cost_per_1m_in": 0.26, - "cost_per_1m_out": 1.04, + "cost_per_1m_in": 0.21, + "cost_per_1m_out": 1.9, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, @@ -2841,12 +2898,12 @@ { "id": "qwen/qwen3-vl-235b-a22b-thinking", "name": "Qwen: Qwen3 VL 235B A22B Thinking", - "cost_per_1m_in": 0, - "cost_per_1m_out": 0, + "cost_per_1m_in": 0.44999999999999996, + "cost_per_1m_out": 3.5, "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": true, "reasoning_levels": [ "low", @@ -2873,12 +2930,12 @@ { "id": "qwen/qwen3-vl-30b-a3b-thinking", "name": "Qwen: Qwen3 VL 30B A3B Thinking", - "cost_per_1m_in": 0, - "cost_per_1m_out": 0, + "cost_per_1m_in": 0.29, + "cost_per_1m_out": 1, "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": true, "reasoning_levels": [ "low", @@ -3185,10 +3242,10 @@ { "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": 0.09999999999999999, + "cost_per_1m_out": 0.3, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.045, + "cost_per_1m_out_cached": 0.09999999999999999, "context_window": 262144, "default_max_tokens": 32768, "can_reason": true, @@ -3222,7 +3279,7 @@ "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0.11, "context_window": 131072, - "default_max_tokens": 48000, + "default_max_tokens": 49152, "can_reason": true, "reasoning_levels": [ "low", @@ -3236,12 +3293,12 @@ { "id": "z-ai/glm-4.5-air", "name": "Z.ai: GLM 4.5 Air", - "cost_per_1m_in": 0.13, - "cost_per_1m_out": 0.85, + "cost_per_1m_in": 0.19999999999999998, + "cost_per_1m_out": 1.2, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.024999999999999998, + "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 49152, + "default_max_tokens": 13107, "can_reason": true, "reasoning_levels": [ "low", @@ -3293,12 +3350,12 @@ { "id": "z-ai/glm-4.6", "name": "Z.ai: GLM 4.6", - "cost_per_1m_in": 0.55, - "cost_per_1m_out": 2.2, + "cost_per_1m_in": 0.43, + "cost_per_1m_out": 1.74, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.11, - "context_window": 204800, - "default_max_tokens": 65536, + "cost_per_1m_out_cached": 0.0799999993, + "context_window": 202752, + "default_max_tokens": 20275, "can_reason": true, "reasoning_levels": [ "low", @@ -3334,9 +3391,9 @@ "cost_per_1m_in": 0.3, "cost_per_1m_out": 0.8999999999999999, "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": 32768, + "default_max_tokens": 65536, "can_reason": true, "reasoning_levels": [ "low", @@ -3350,12 +3407,12 @@ { "id": "z-ai/glm-4.7", "name": "Z.ai: GLM 4.7", - "cost_per_1m_in": 0.54, - "cost_per_1m_out": 1.9800000000000002, + "cost_per_1m_in": 0.44999999999999996, + "cost_per_1m_out": 2.2, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.099, + "cost_per_1m_out_cached": 0.11, "context_window": 204800, - "default_max_tokens": 65536, + "default_max_tokens": 102400, "can_reason": true, "reasoning_levels": [ "low",