diff --git a/internal/providers/configs/openrouter.json b/internal/providers/configs/openrouter.json index dbb2bcbad16a9fb248c9859ef3615d28dc276f58..4fac1e065b6d13a0c18c410e1e4e7efe54a40d6e 100644 --- a/internal/providers/configs/openrouter.json +++ b/internal/providers/configs/openrouter.json @@ -158,7 +158,7 @@ "cost_per_1m_in_cached": 3.75, "cost_per_1m_out_cached": 0.3, "context_window": 200000, - "default_max_tokens": 32000, + "default_max_tokens": 64000, "can_reason": true, "reasoning_levels": [ "low", @@ -427,8 +427,8 @@ { "id": "deepseek/deepseek-chat-v3-0324", "name": "DeepSeek: DeepSeek V3 0324", - "cost_per_1m_in": 0.28, - "cost_per_1m_out": 0.88, + "cost_per_1m_in": 0.9, + "cost_per_1m_out": 0.9, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 163840, @@ -495,8 +495,8 @@ "cost_per_1m_out": 1, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, - "context_window": 131072, - "default_max_tokens": 32768, + "context_window": 163840, + "default_max_tokens": 16384, "can_reason": true, "reasoning_levels": [ "low", @@ -647,7 +647,7 @@ "cost_per_1m_in": 0.09999999999999999, "cost_per_1m_out": 0.39999999999999997, "cost_per_1m_in_cached": 0.18330000000000002, - "cost_per_1m_out_cached": 0.01, + "cost_per_1m_out_cached": 0.024999999999999998, "context_window": 1048576, "default_max_tokens": 32767, "can_reason": true, @@ -687,7 +687,7 @@ "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 1048576, - "default_max_tokens": 32768, + "default_max_tokens": 32767, "can_reason": true, "reasoning_levels": [ "low", @@ -706,7 +706,7 @@ "cost_per_1m_in_cached": 0.3833, "cost_per_1m_out_cached": 0.075, "context_window": 1048576, - "default_max_tokens": 32767, + "default_max_tokens": 32768, "can_reason": true, "reasoning_levels": [ "low", @@ -868,12 +868,12 @@ { "id": "meta-llama/llama-3.1-8b-instruct", "name": "Meta: Llama 3.1 8B Instruct", - "cost_per_1m_in": 0.049999999999999996, - "cost_per_1m_out": 0.08, + "cost_per_1m_in": 0.03, + "cost_per_1m_out": 0.049999999999999996, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 65536, + "default_max_tokens": 8192, "can_reason": false, "supports_attachments": false, "options": {} @@ -895,11 +895,11 @@ "id": "meta-llama/llama-3.3-70b-instruct", "name": "Meta: Llama 3.3 70B Instruct", "cost_per_1m_in": 0.13, - "cost_per_1m_out": 0.39999999999999997, + "cost_per_1m_out": 0.38, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.07, + "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 65536, + "default_max_tokens": 8192, "can_reason": false, "supports_attachments": false, "options": {} @@ -985,12 +985,12 @@ { "id": "minimax/minimax-m2", "name": "MiniMax: MiniMax M2", - "cost_per_1m_in": 0.3, - "cost_per_1m_out": 1.2, + "cost_per_1m_in": 0.27, + "cost_per_1m_out": 1.15, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, - "context_window": 204800, - "default_max_tokens": 65536, + "context_window": 262144, + "default_max_tokens": 26214, "can_reason": true, "reasoning_levels": [ "low", @@ -1399,12 +1399,12 @@ { "id": "moonshotai/kimi-k2", "name": "MoonshotAI: Kimi K2 0711", - "cost_per_1m_in": 0.5, - "cost_per_1m_out": 2.4, + "cost_per_1m_in": 0.5700000000000001, + "cost_per_1m_out": 2.3, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 13107, + "default_max_tokens": 65536, "can_reason": false, "supports_attachments": false, "options": {} @@ -1438,7 +1438,7 @@ { "id": "moonshotai/kimi-k2-thinking", "name": "MoonshotAI: Kimi K2 Thinking", - "cost_per_1m_in": 0.55, + "cost_per_1m_in": 0.6, "cost_per_1m_out": 2.5, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, @@ -1660,7 +1660,7 @@ "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, "options": {} @@ -1673,7 +1673,7 @@ "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0.09999999999999999, "context_window": 1047576, - "default_max_tokens": 16384, + "default_max_tokens": 104757, "can_reason": false, "supports_attachments": true, "options": {} @@ -1896,9 +1896,9 @@ "cost_per_1m_in": 0.049999999999999996, "cost_per_1m_out": 0.39999999999999997, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.005, + "cost_per_1m_out_cached": 0.01, "context_window": 400000, - "default_max_tokens": 64000, + "default_max_tokens": 40000, "can_reason": true, "reasoning_levels": [ "low", @@ -2236,12 +2236,12 @@ { "id": "qwen/qwen-2.5-72b-instruct", "name": "Qwen2.5 72B Instruct", - "cost_per_1m_in": 0.12, - "cost_per_1m_out": 0.39, + "cost_per_1m_in": 0.07, + "cost_per_1m_out": 0.26, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 32768, - "default_max_tokens": 8192, + "default_max_tokens": 16384, "can_reason": false, "supports_attachments": false, "options": {} @@ -2333,8 +2333,8 @@ { "id": "qwen/qwen3-14b", "name": "Qwen: Qwen3 14B", - "cost_per_1m_in": 0.08, - "cost_per_1m_out": 0.24, + "cost_per_1m_in": 0.049999999999999996, + "cost_per_1m_out": 0.22, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 40960, @@ -2428,8 +2428,8 @@ { "id": "qwen/qwen3-30b-a3b", "name": "Qwen: Qwen3 30B A3B", - "cost_per_1m_in": 0.08, - "cost_per_1m_out": 0.28, + "cost_per_1m_in": 0.09, + "cost_per_1m_out": 0.44999999999999996, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, @@ -2447,12 +2447,12 @@ { "id": "qwen/qwen3-30b-a3b-instruct-2507", "name": "Qwen: Qwen3 30B A3B Instruct 2507", - "cost_per_1m_in": 0.09999999999999999, - "cost_per_1m_out": 0.3, + "cost_per_1m_in": 0.08, + "cost_per_1m_out": 0.33, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 26214, + "default_max_tokens": 131072, "can_reason": false, "supports_attachments": false, "options": {} @@ -2549,8 +2549,8 @@ { "id": "qwen/qwen3-coder", "name": "Qwen: Qwen3 Coder 480B A35B", - "cost_per_1m_in": 1, - "cost_per_1m_out": 1.5, + "cost_per_1m_in": 0.25, + "cost_per_1m_out": 1, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 262144, @@ -2562,19 +2562,13 @@ { "id": "qwen/qwen3-coder:exacto", "name": "Qwen: Qwen3 Coder 480B A35B (exacto)", - "cost_per_1m_in": 0.38, - "cost_per_1m_out": 1.53, + "cost_per_1m_in": 1, + "cost_per_1m_out": 4, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 131072, - "can_reason": true, - "reasoning_levels": [ - "low", - "medium", - "high" - ], - "default_reasoning_effort": "medium", + "default_max_tokens": 16384, + "can_reason": false, "supports_attachments": false, "options": {} }, @@ -2665,12 +2659,12 @@ { "id": "qwen/qwen3-vl-235b-a22b-instruct", "name": "Qwen: Qwen3 VL 235B A22B Instruct", - "cost_per_1m_in": 0.3, - "cost_per_1m_out": 1.2, + "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": 131072, + "default_max_tokens": 26214, "can_reason": false, "supports_attachments": true, "options": {} @@ -2697,12 +2691,12 @@ { "id": "qwen/qwen3-vl-30b-a3b-instruct", "name": "Qwen: Qwen3 VL 30B A3B Instruct", - "cost_per_1m_in": 0.15, - "cost_per_1m_out": 0.6, + "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": 262144, - "default_max_tokens": 26214, + "default_max_tokens": 131072, "can_reason": false, "supports_attachments": true, "options": {} @@ -2729,12 +2723,12 @@ { "id": "qwen/qwen3-vl-8b-instruct", "name": "Qwen: Qwen3 VL 8B Instruct", - "cost_per_1m_in": 0.12, - "cost_per_1m_out": 0.7, + "cost_per_1m_in": 0.08, + "cost_per_1m_out": 0.5, "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": false, "supports_attachments": true, "options": {} @@ -2858,7 +2852,7 @@ "id": "alibaba/tongyi-deepresearch-30b-a3b", "name": "Tongyi DeepResearch 30B A3B", "cost_per_1m_in": 0.09, - "cost_per_1m_out": 0.39999999999999997, + "cost_per_1m_out": 0.44999999999999996, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, @@ -2985,11 +2979,11 @@ "id": "z-ai/glm-4.6", "name": "Z.AI: GLM 4.6", "cost_per_1m_in": 0.6, - "cost_per_1m_out": 2.2, + "cost_per_1m_out": 2, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.11, + "cost_per_1m_out_cached": 0, "context_window": 204800, - "default_max_tokens": 65536, + "default_max_tokens": 20480, "can_reason": true, "reasoning_levels": [ "low", @@ -3054,10 +3048,10 @@ { "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, @@ -3067,10 +3061,10 @@ { "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, @@ -3080,10 +3074,10 @@ { "id": "x-ai/grok-3-mini", "name": "xAI: Grok 3 Mini", - "cost_per_1m_in": 0.3, - "cost_per_1m_out": 0.5, + "cost_per_1m_in": 0.6, + "cost_per_1m_out": 4, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.075, + "cost_per_1m_out_cached": 0.15, "context_window": 131072, "default_max_tokens": 13107, "can_reason": true, @@ -3099,10 +3093,10 @@ { "id": "x-ai/grok-3-mini-beta", "name": "xAI: Grok 3 Mini Beta", - "cost_per_1m_in": 0.3, - "cost_per_1m_out": 0.5, + "cost_per_1m_in": 0.6, + "cost_per_1m_out": 4, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.075, + "cost_per_1m_out_cached": 0.15, "context_window": 131072, "default_max_tokens": 13107, "can_reason": true,