diff --git a/internal/providers/configs/openrouter.json b/internal/providers/configs/openrouter.json index 496f5e4ac0033a16488cf5a3afab963f295de086..378697cd5d92d55503b46f328d9f65475ee1b516 100644 --- a/internal/providers/configs/openrouter.json +++ b/internal/providers/configs/openrouter.json @@ -171,7 +171,7 @@ "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, "has_reasoning_efforts": true, "supports_attachments": true @@ -374,12 +374,12 @@ { "id": "deepseek/deepseek-chat-v3-0324", "name": "DeepSeek: DeepSeek V3 0324", - "cost_per_1m_in": 0.5, - "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": 163840, - "default_max_tokens": 16384, + "default_max_tokens": 81920, "can_reason": false, "has_reasoning_efforts": false, "supports_attachments": false @@ -400,12 +400,12 @@ { "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": 0.27, + "cost_per_1m_out": 1, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 163840, - "default_max_tokens": 10240, + "default_max_tokens": 16384, "can_reason": true, "has_reasoning_efforts": true, "supports_attachments": false @@ -423,11 +423,24 @@ "has_reasoning_efforts": true, "supports_attachments": false }, + { + "id": "deepseek/deepseek-v3.1-terminus:exacto", + "name": "DeepSeek: DeepSeek V3.1 Terminus (exacto)", + "cost_per_1m_in": 0.27, + "cost_per_1m_out": 1, + "cost_per_1m_in_cached": 0, + "cost_per_1m_out_cached": 0, + "context_window": 131072, + "default_max_tokens": 32768, + "can_reason": true, + "has_reasoning_efforts": true, + "supports_attachments": false + }, { "id": "deepseek/deepseek-v3.2-exp", "name": "DeepSeek: DeepSeek V3.2 Exp", - "cost_per_1m_in": 0.28, - "cost_per_1m_out": 0.39999999999999997, + "cost_per_1m_in": 0.27, + "cost_per_1m_out": 0.41, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 163840, @@ -452,12 +465,12 @@ { "id": "deepseek/deepseek-r1-0528", "name": "DeepSeek: R1 0528", - "cost_per_1m_in": 0.7999999999999999, - "cost_per_1m_out": 2.4, + "cost_per_1m_in": 1.35, + "cost_per_1m_out": 5.4, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 163840, - "default_max_tokens": 16384, + "default_max_tokens": 10240, "can_reason": true, "has_reasoning_efforts": true, "supports_attachments": false @@ -618,19 +631,6 @@ "has_reasoning_efforts": true, "supports_attachments": true }, - { - "id": "google/gemma-3-27b-it", - "name": "Google: Gemma 3 27B", - "cost_per_1m_in": 0.13, - "cost_per_1m_out": 0.52, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, - "context_window": 96000, - "default_max_tokens": 48000, - "can_reason": false, - "has_reasoning_efforts": false, - "supports_attachments": true - }, { "id": "inception/mercury", "name": "Inception: Mercury", @@ -712,12 +712,12 @@ { "id": "meta-llama/llama-3.1-70b-instruct", "name": "Meta: Llama 3.1 70B Instruct", - "cost_per_1m_in": 0.88, - "cost_per_1m_out": 0.88, + "cost_per_1m_in": 0.39999999999999997, + "cost_per_1m_out": 0.39999999999999997, "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, "has_reasoning_efforts": false, "supports_attachments": false @@ -751,12 +751,12 @@ { "id": "meta-llama/llama-3.3-70b-instruct", "name": "Meta: Llama 3.3 70B Instruct", - "cost_per_1m_in": 0.25, - "cost_per_1m_out": 0.75, + "cost_per_1m_in": 0.59, + "cost_per_1m_out": 0.7899999999999999, "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, "has_reasoning_efforts": false, "supports_attachments": false @@ -790,12 +790,12 @@ { "id": "meta-llama/llama-4-maverick", "name": "Meta: Llama 4 Maverick", - "cost_per_1m_in": 0.22, - "cost_per_1m_out": 0.88, + "cost_per_1m_in": 0.16999999999999998, + "cost_per_1m_out": 0.85, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 1048576, - "default_max_tokens": 65536, + "default_max_tokens": 524288, "can_reason": false, "has_reasoning_efforts": false, "supports_attachments": true @@ -1154,12 +1154,12 @@ { "id": "mistralai/mistral-small-3.1-24b-instruct", "name": "Mistral: Mistral Small 3.1 24B", - "cost_per_1m_in": 0.09999999999999999, - "cost_per_1m_out": 0.3, + "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": 131072, - "default_max_tokens": 13107, + "default_max_tokens": 65536, "can_reason": false, "has_reasoning_efforts": false, "supports_attachments": true @@ -1180,12 +1180,12 @@ { "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, "context_window": 131072, - "default_max_tokens": 65536, + "default_max_tokens": 13107, "can_reason": false, "has_reasoning_efforts": false, "supports_attachments": true @@ -1271,12 +1271,12 @@ { "id": "moonshotai/kimi-k2", "name": "MoonshotAI: Kimi K2 0711", - "cost_per_1m_in": 0.6, - "cost_per_1m_out": 2.5, + "cost_per_1m_in": 0.5, + "cost_per_1m_out": 2.4, "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, "has_reasoning_efforts": false, "supports_attachments": false @@ -1284,8 +1284,21 @@ { "id": "moonshotai/kimi-k2-0905", "name": "MoonshotAI: Kimi K2 0905", - "cost_per_1m_in": 0.5, - "cost_per_1m_out": 2, + "cost_per_1m_in": 1, + "cost_per_1m_out": 3, + "cost_per_1m_in_cached": 0, + "cost_per_1m_out_cached": 0, + "context_window": 262144, + "default_max_tokens": 26214, + "can_reason": false, + "has_reasoning_efforts": false, + "supports_attachments": false + }, + { + "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_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 262144, @@ -1323,8 +1336,8 @@ { "id": "nvidia/nemotron-nano-9b-v2", "name": "NVIDIA: Nemotron Nano 9B V2", - "cost_per_1m_in": 0.06, - "cost_per_1m_out": 0.25, + "cost_per_1m_in": 0.04, + "cost_per_1m_out": 0.16, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, @@ -1385,6 +1398,19 @@ "has_reasoning_efforts": false, "supports_attachments": false }, + { + "id": "nousresearch/hermes-4-405b", + "name": "Nous: Hermes 4 405B", + "cost_per_1m_in": 0.3, + "cost_per_1m_out": 1.2, + "cost_per_1m_in_cached": 0, + "cost_per_1m_out_cached": 0, + "context_window": 131072, + "default_max_tokens": 65536, + "can_reason": true, + "has_reasoning_efforts": true, + "supports_attachments": false + }, { "id": "nousresearch/hermes-4-70b", "name": "Nous: Hermes 4 70B", @@ -1752,12 +1778,25 @@ { "id": "openai/gpt-oss-120b", "name": "OpenAI: gpt-oss-120b", - "cost_per_1m_in": 0.04, - "cost_per_1m_out": 0.39999999999999997, + "cost_per_1m_in": 0.09999999999999999, + "cost_per_1m_out": 0.49, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 65536, + "default_max_tokens": 13107, + "can_reason": true, + "has_reasoning_efforts": true, + "supports_attachments": false + }, + { + "id": "openai/gpt-oss-120b:exacto", + "name": "OpenAI: gpt-oss-120b (exacto)", + "cost_per_1m_in": 0.049999999999999996, + "cost_per_1m_out": 0.24, + "cost_per_1m_in_cached": 0, + "cost_per_1m_out_cached": 0, + "context_window": 131072, + "default_max_tokens": 13107, "can_reason": true, "has_reasoning_efforts": true, "supports_attachments": false @@ -1918,19 +1957,6 @@ "has_reasoning_efforts": false, "supports_attachments": false }, - { - "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, - "has_reasoning_efforts": true, - "supports_attachments": false - }, { "id": "qwen/qwen-plus-2025-07-28", "name": "Qwen: Qwen Plus 0728", @@ -2012,12 +2038,12 @@ { "id": "qwen/qwen3-14b", "name": "Qwen: Qwen3 14B", - "cost_per_1m_in": 0.08, + "cost_per_1m_in": 0.06, "cost_per_1m_out": 0.24, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 40960, - "default_max_tokens": 4096, + "default_max_tokens": 20480, "can_reason": true, "has_reasoning_efforts": true, "supports_attachments": false @@ -2025,12 +2051,12 @@ { "id": "qwen/qwen3-235b-a22b", "name": "Qwen: Qwen3 235B A22B", - "cost_per_1m_in": 0.22, - "cost_per_1m_out": 0.88, + "cost_per_1m_in": 0.3, + "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, + "context_window": 40960, + "default_max_tokens": 20480, "can_reason": true, "has_reasoning_efforts": true, "supports_attachments": false @@ -2044,19 +2070,19 @@ "cost_per_1m_out_cached": 0, "context_window": 131072, "default_max_tokens": 13107, - "can_reason": false, - "has_reasoning_efforts": false, + "can_reason": true, + "has_reasoning_efforts": true, "supports_attachments": false }, { "id": "qwen/qwen3-235b-a22b-2507", "name": "Qwen: Qwen3 235B A22B Instruct 2507", - "cost_per_1m_in": 0.19999999999999998, - "cost_per_1m_out": 0.6, + "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, - "default_max_tokens": 26214, + "default_max_tokens": 8192, "can_reason": false, "has_reasoning_efforts": false, "supports_attachments": false @@ -2090,12 +2116,12 @@ { "id": "qwen/qwen3-30b-a3b-instruct-2507", "name": "Qwen: Qwen3 30B A3B Instruct 2507", - "cost_per_1m_in": 0.09, + "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, "has_reasoning_efforts": false, "supports_attachments": false @@ -2103,12 +2129,12 @@ { "id": "qwen/qwen3-30b-a3b-thinking-2507", "name": "Qwen: Qwen3 30B A3B Thinking 2507", - "cost_per_1m_in": 0.08, - "cost_per_1m_out": 0.29, + "cost_per_1m_in": 0.09999999999999999, + "cost_per_1m_out": 0.3, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 131072, + "default_max_tokens": 26214, "can_reason": true, "has_reasoning_efforts": true, "supports_attachments": false @@ -2116,12 +2142,12 @@ { "id": "qwen/qwen3-32b", "name": "Qwen: Qwen3 32B", - "cost_per_1m_in": 0.15, - "cost_per_1m_out": 0.5, + "cost_per_1m_in": 0.14, + "cost_per_1m_out": 0.5700000000000001, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 131072, - "default_max_tokens": 4000, + "default_max_tokens": 65536, "can_reason": true, "has_reasoning_efforts": true, "supports_attachments": false @@ -2142,12 +2168,12 @@ { "id": "qwen/qwen3-coder-30b-a3b-instruct", "name": "Qwen: Qwen3 Coder 30B A3B Instruct", - "cost_per_1m_in": 0.06, - "cost_per_1m_out": 0.25, + "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": 262144, - "default_max_tokens": 131072, + "default_max_tokens": 26214, "can_reason": false, "has_reasoning_efforts": false, "supports_attachments": false @@ -2155,14 +2181,27 @@ { "id": "qwen/qwen3-coder", "name": "Qwen: Qwen3 Coder 480B A35B", - "cost_per_1m_in": 0.29, - "cost_per_1m_out": 1.2, + "cost_per_1m_in": 0.38, + "cost_per_1m_out": 1.53, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 32768, - "can_reason": false, - "has_reasoning_efforts": false, + "default_max_tokens": 131072, + "can_reason": true, + "has_reasoning_efforts": true, + "supports_attachments": false + }, + { + "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_cached": 0, + "cost_per_1m_out_cached": 0, + "context_window": 262144, + "default_max_tokens": 131072, + "can_reason": true, + "has_reasoning_efforts": true, "supports_attachments": false }, { @@ -2233,12 +2272,12 @@ { "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": 0.14, + "cost_per_1m_out": 1.2, "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, "has_reasoning_efforts": true, "supports_attachments": false @@ -2246,12 +2285,12 @@ { "id": "qwen/qwen3-vl-235b-a22b-instruct", "name": "Qwen: Qwen3 VL 235B A22B Instruct", - "cost_per_1m_in": 0.5, - "cost_per_1m_out": 2.5, + "cost_per_1m_in": 0.3, + "cost_per_1m_out": 1.2, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, - "context_window": 131072, - "default_max_tokens": 65536, + "context_window": 262144, + "default_max_tokens": 131072, "can_reason": false, "has_reasoning_efforts": false, "supports_attachments": true @@ -2272,12 +2311,12 @@ { "id": "qwen/qwen3-vl-30b-a3b-instruct", "name": "Qwen: Qwen3 VL 30B A3B Instruct", - "cost_per_1m_in": 0.19999999999999998, - "cost_per_1m_out": 0.7, + "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": false, "has_reasoning_efforts": false, "supports_attachments": true @@ -2390,7 +2429,7 @@ "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, @@ -2441,12 +2480,12 @@ { "id": "z-ai/glm-4.5-air", "name": "Z.AI: GLM 4.5 Air", - "cost_per_1m_in": 0.19999999999999998, - "cost_per_1m_out": 1.1, + "cost_per_1m_in": 0.13, + "cost_per_1m_out": 0.85, "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": 48000, + "default_max_tokens": 49152, "can_reason": true, "has_reasoning_efforts": true, "supports_attachments": false @@ -2490,6 +2529,19 @@ "has_reasoning_efforts": true, "supports_attachments": false }, + { + "id": "z-ai/glm-4.6:exacto", + "name": "Z.AI: GLM 4.6 (exacto)", + "cost_per_1m_in": 0.6, + "cost_per_1m_out": 1.9, + "cost_per_1m_in_cached": 0, + "cost_per_1m_out_cached": 0, + "context_window": 202752, + "default_max_tokens": 20275, + "can_reason": true, + "has_reasoning_efforts": true, + "supports_attachments": false + }, { "id": "inclusionai/ling-1t", "name": "inclusionAI: Ling-1T", @@ -2519,10 +2571,10 @@ { "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, @@ -2532,10 +2584,10 @@ { "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, @@ -2545,10 +2597,10 @@ { "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, @@ -2558,10 +2610,10 @@ { "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,