diff --git a/internal/providers/configs/cortecs.json b/internal/providers/configs/cortecs.json index 02de18f8aea8df486bc1fd85c7bf60a864d3f49c..fc9cdae366f106e0a7fa190a15c96137ad3bc4fd 100644 --- a/internal/providers/configs/cortecs.json +++ b/internal/providers/configs/cortecs.json @@ -340,12 +340,12 @@ { "id": "glm-4.7", "name": "GLM 4.7", - "cost_per_1m_in": 0.376, - "cost_per_1m_out": 1.878, + "cost_per_1m_in": 0.559, + "cost_per_1m_out": 2.05, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, - "context_window": 200000, - "default_max_tokens": 20000, + "context_window": 202752, + "default_max_tokens": 20275, "can_reason": true, "reasoning_levels": [ "low", @@ -358,8 +358,8 @@ { "id": "minimax-m2.1", "name": "MiniMax M2.1", - "cost_per_1m_in": 0.282, - "cost_per_1m_out": 1.127, + "cost_per_1m_in": 0.338, + "cost_per_1m_out": 1.352, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 196000, @@ -373,18 +373,6 @@ "default_reasoning_effort": "medium", "supports_attachments": false }, - { - "id": "llama-guard-3-8b", - "name": "Llama Guard 3 8B", - "cost_per_1m_in": 0.019, - "cost_per_1m_out": 0.056, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, - "context_window": 128000, - "default_max_tokens": 12800, - "can_reason": false, - "supports_attachments": false - }, { "id": "qwen3-vl-235b-a22b", "name": "Qwen3 VL 235B A22B", @@ -583,24 +571,6 @@ "can_reason": false, "supports_attachments": true }, - { - "id": "kimi-k2-thinking", - "name": "Kimi K2 Thinking", - "cost_per_1m_in": 0.564, - "cost_per_1m_out": 2.348, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, - "context_window": 262000, - "default_max_tokens": 26200, - "can_reason": true, - "reasoning_levels": [ - "low", - "medium", - "high" - ], - "default_reasoning_effort": "medium", - "supports_attachments": false - }, { "id": "intellect-3", "name": "INTELLECT-3", @@ -640,8 +610,8 @@ { "id": "nemotron-nano-v2-12b", "name": "Nemotron Nano V2 12b", - "cost_per_1m_in": 0.066, - "cost_per_1m_out": 0.188, + "cost_per_1m_in": 0.226, + "cost_per_1m_out": 0.667, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 128000, @@ -757,24 +727,6 @@ "default_reasoning_effort": "medium", "supports_attachments": true }, - { - "id": "qwen3-235b-a22b-thinking-2507", - "name": "Qwen3 235B A22B Thinking 2507", - "cost_per_1m_in": 0.188, - "cost_per_1m_out": 0.752, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, - "context_window": 262000, - "default_max_tokens": 26200, - "can_reason": true, - "reasoning_levels": [ - "low", - "medium", - "high" - ], - "default_reasoning_effort": "medium", - "supports_attachments": false - }, { "id": "gpt-oss-120b", "name": "GPT Oss 120b", @@ -811,42 +763,6 @@ "default_reasoning_effort": "medium", "supports_attachments": false }, - { - "id": "qwen3-30b-a3b-thinking-2507", - "name": "Qwen3 30B A3B Thinking 2507", - "cost_per_1m_in": 0.093, - "cost_per_1m_out": 0.281, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, - "context_window": 262000, - "default_max_tokens": 26200, - "can_reason": true, - "reasoning_levels": [ - "low", - "medium", - "high" - ], - "default_reasoning_effort": "medium", - "supports_attachments": false - }, - { - "id": "qwen3-coder-480b-a35b-instruct", - "name": "Qwen3 Coder 480B A35B Instruct", - "cost_per_1m_in": 0.376, - "cost_per_1m_out": 1.691, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, - "context_window": 262000, - "default_max_tokens": 26200, - "can_reason": true, - "reasoning_levels": [ - "low", - "medium", - "high" - ], - "default_reasoning_effort": "medium", - "supports_attachments": false - }, { "id": "gpt-oss-20b", "name": "GPT Oss 20b", @@ -865,54 +781,6 @@ "default_reasoning_effort": "medium", "supports_attachments": false }, - { - "id": "kimi-k2-instruct", - "name": "Kimi K2 Instruct", - "cost_per_1m_in": 0.469, - "cost_per_1m_out": 2.254, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, - "context_window": 131000, - "default_max_tokens": 13100, - "can_reason": false, - "supports_attachments": false - }, - { - "id": "glm-4.5", - "name": "GLM 4.5", - "cost_per_1m_in": 0.564, - "cost_per_1m_out": 2.066, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, - "context_window": 128000, - "default_max_tokens": 12800, - "can_reason": true, - "reasoning_levels": [ - "low", - "medium", - "high" - ], - "default_reasoning_effort": "medium", - "supports_attachments": false - }, - { - "id": "glm-4.5-air", - "name": "GLM 4.5 Air", - "cost_per_1m_in": 0.188, - "cost_per_1m_out": 1.127, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, - "context_window": 128000, - "default_max_tokens": 12800, - "can_reason": true, - "reasoning_levels": [ - "low", - "medium", - "high" - ], - "default_reasoning_effort": "medium", - "supports_attachments": false - }, { "id": "mistral-7b-instruct-v0.3", "name": "Mistral 7B Instruct v0.3", @@ -1209,7 +1077,7 @@ "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 163840, - "default_max_tokens": 12800, + "default_max_tokens": 16384, "can_reason": true, "reasoning_levels": [ "low", @@ -1342,8 +1210,8 @@ { "id": "deepseek-r1-0528", "name": "DeepSeek R1 0528", - "cost_per_1m_in": 0.752, - "cost_per_1m_out": 2.254, + "cost_per_1m_in": 0.614, + "cost_per_1m_out": 2.423, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, "context_window": 164000, diff --git a/internal/providers/configs/huggingface.json b/internal/providers/configs/huggingface.json index ffa408f079ab5c5e99be44adbbe2c13021dceceb..c6a54815f59e331bbc8d39630019d3376d48b4a1 100644 --- a/internal/providers/configs/huggingface.json +++ b/internal/providers/configs/huggingface.json @@ -8,8 +8,8 @@ "default_small_model_id": "openai/gpt-oss-20b:groq", "models": [ { - "id": "MiniMaxAI/MiniMax-M2.5:fireworks-ai", - "name": "MiniMaxAI/MiniMax-M2.5 (fireworks-ai)", + "id": "MiniMaxAI/MiniMax-M2.1:fireworks-ai", + "name": "MiniMaxAI/MiniMax-M2.1 (fireworks-ai)", "cost_per_1m_in": 0, "cost_per_1m_out": 0, "cost_per_1m_in_cached": 0, @@ -26,7 +26,7 @@ "cost_per_1m_out": 0, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, - "context_window": 204800, + "context_window": 196608, "default_max_tokens": 8192, "can_reason": false, "supports_attachments": false @@ -38,7 +38,7 @@ "cost_per_1m_out": 1.2, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, - "context_window": 131072, + "context_window": 32768, "default_max_tokens": 8192, "can_reason": false, "supports_attachments": false @@ -91,18 +91,6 @@ "can_reason": false, "supports_attachments": false }, - { - "id": "deepcogito/cogito-671b-v2.1:fireworks-ai", - "name": "deepcogito/cogito-671b-v2.1 (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": 163840, - "default_max_tokens": 8192, - "can_reason": false, - "supports_attachments": false - }, { "id": "meta-llama/Llama-3.3-70B-Instruct:groq", "name": "meta-llama/Llama-3.3-70B-Instruct (groq)", @@ -151,18 +139,6 @@ "can_reason": false, "supports_attachments": false }, - { - "id": "moonshotai/Kimi-K2-Thinking:fireworks-ai", - "name": "moonshotai/Kimi-K2-Thinking (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": 262144, - "default_max_tokens": 8192, - "can_reason": false, - "supports_attachments": false - }, { "id": "moonshotai/Kimi-K2.5:fireworks-ai", "name": "moonshotai/Kimi-K2.5 (fireworks-ai)", @@ -254,7 +230,7 @@ "cost_per_1m_out": 0, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, - "context_window": 204800, + "context_window": 202752, "default_max_tokens": 8192, "can_reason": false, "supports_attachments": false @@ -294,6 +270,18 @@ "default_max_tokens": 8192, "can_reason": false, "supports_attachments": false + }, + { + "id": "zai-org/GLM-5.1-FP8:fireworks-ai", + "name": "zai-org/GLM-5.1-FP8 (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 } ], "default_headers": { diff --git a/internal/providers/configs/ionet.json b/internal/providers/configs/ionet.json index beeadd9bc49ea0a87fd686b2486cce639afd5bb8..a9af6b204d544fe2fe7eadfd2d7ce4de45f2f8b4 100644 --- a/internal/providers/configs/ionet.json +++ b/internal/providers/configs/ionet.json @@ -119,7 +119,7 @@ "high" ], "default_reasoning_effort": "medium", - "supports_attachments": false + "supports_attachments": true }, { "id": "openai/gpt-oss-120b", diff --git a/internal/providers/configs/openrouter.json b/internal/providers/configs/openrouter.json index cb6e2e274c861f02fb60ee30c8727e2d7150514b..efd86772229faf8ccb3c5423be573cd7e5b93f78 100644 --- a/internal/providers/configs/openrouter.json +++ b/internal/providers/configs/openrouter.json @@ -532,12 +532,12 @@ { "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": 0.27, + "cost_per_1m_out": 1, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.135, "context_window": 163840, - "default_max_tokens": 16384, + "default_max_tokens": 32768, "can_reason": true, "reasoning_levels": [ "low", @@ -589,9 +589,9 @@ "cost_per_1m_in": 0.27, "cost_per_1m_out": 0.41, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.27, "context_window": 163840, - "default_max_tokens": 32768, + "default_max_tokens": 81920, "can_reason": true, "reasoning_levels": [ "low", @@ -868,12 +868,12 @@ { "id": "google/gemma-4-26b-a4b-it", "name": "Google: Gemma 4 26B A4B ", - "cost_per_1m_in": 0.13, - "cost_per_1m_out": 0.4, + "cost_per_1m_in": 0.08, + "cost_per_1m_out": 0.35, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, + "cost_per_1m_out_cached": 0.01, "context_window": 262144, - "default_max_tokens": 65536, + "default_max_tokens": 26214, "can_reason": true, "reasoning_levels": [ "low", @@ -937,18 +937,6 @@ "default_reasoning_effort": "medium", "supports_attachments": true }, - { - "id": "inception/mercury", - "name": "Inception: Mercury", - "cost_per_1m_in": 0.25, - "cost_per_1m_out": 0.75, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.025, - "context_window": 128000, - "default_max_tokens": 16000, - "can_reason": false, - "supports_attachments": false - }, { "id": "inception/mercury-2", "name": "Inception: Mercury 2", @@ -967,18 +955,6 @@ "default_reasoning_effort": "medium", "supports_attachments": false }, - { - "id": "inception/mercury-coder", - "name": "Inception: Mercury Coder", - "cost_per_1m_in": 0.25, - "cost_per_1m_out": 0.75, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.025, - "context_window": 128000, - "default_max_tokens": 16000, - "can_reason": false, - "supports_attachments": false - }, { "id": "kwaipilot/kat-coder-pro-v2", "name": "Kwaipilot: KAT-Coder-Pro V2", @@ -986,20 +962,8 @@ "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, - "can_reason": false, - "supports_attachments": false - }, - { - "id": "meituan/longcat-flash-chat", - "name": "Meituan: LongCat Flash Chat", - "cost_per_1m_in": 0.2, - "cost_per_1m_out": 0.8, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.2, - "context_window": 131072, - "default_max_tokens": 65536, + "context_window": 262144, + "default_max_tokens": 72000, "can_reason": false, "supports_attachments": false }, @@ -1051,33 +1015,15 @@ "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, + "cost_per_1m_out": 1.02, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.03, - "context_window": 196608, - "default_max_tokens": 98304, + "cost_per_1m_out_cached": 0, + "context_window": 204800, + "default_max_tokens": 65536, "can_reason": true, "reasoning_levels": [ "low", @@ -1093,9 +1039,9 @@ "cost_per_1m_in": 0.3, "cost_per_1m_out": 1.2, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, - "context_window": 196608, - "default_max_tokens": 19660, + "cost_per_1m_out_cached": 0.03, + "context_window": 204800, + "default_max_tokens": 65536, "can_reason": true, "reasoning_levels": [ "low", @@ -1113,7 +1059,7 @@ "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", @@ -1270,12 +1216,12 @@ { "id": "mistralai/ministral-8b-2512", "name": "Mistral: Ministral 3 8B 2512", - "cost_per_1m_in": 0.15, - "cost_per_1m_out": 0.15, + "cost_per_1m_in": 0.3, + "cost_per_1m_out": 0.3, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.015, + "cost_per_1m_out_cached": 0, "context_window": 262144, - "default_max_tokens": 26214, + "default_max_tokens": 131072, "can_reason": false, "supports_attachments": true }, @@ -1444,12 +1390,12 @@ { "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, + "cost_per_1m_out_cached": 0.5, "context_window": 262144, - "default_max_tokens": 131072, + "default_max_tokens": 8192, "can_reason": false, "supports_attachments": false }, @@ -1459,7 +1405,7 @@ "cost_per_1m_in": 0.6, "cost_per_1m_out": 2.5, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.15, + "cost_per_1m_out_cached": 0, "context_window": 262144, "default_max_tokens": 131072, "can_reason": true, @@ -1474,10 +1420,10 @@ { "id": "moonshotai/kimi-k2.5", "name": "MoonshotAI: Kimi K2.5", - "cost_per_1m_in": 0.5, - "cost_per_1m_out": 2.8, + "cost_per_1m_in": 0.45, + "cost_per_1m_out": 2.25, "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": 26214, "can_reason": true, @@ -1747,9 +1693,9 @@ "cost_per_1m_in": 0.1, "cost_per_1m_out": 0.4, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.025, + "cost_per_1m_out_cached": 0.03, "context_window": 1047576, - "default_max_tokens": 16384, + "default_max_tokens": 104757, "can_reason": false, "supports_attachments": true }, @@ -1927,9 +1873,9 @@ "cost_per_1m_in": 0.25, "cost_per_1m_out": 2, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.025, + "cost_per_1m_out_cached": 0.03, "context_window": 400000, - "default_max_tokens": 64000, + "default_max_tokens": 40000, "can_reason": true, "reasoning_levels": [ "low", @@ -1945,9 +1891,9 @@ "cost_per_1m_in": 0.05, "cost_per_1m_out": 0.4, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.005, + "cost_per_1m_out_cached": 0.01, "context_window": 400000, - "default_max_tokens": 64000, + "default_max_tokens": 40000, "can_reason": true, "reasoning_levels": [ "low", @@ -2230,12 +2176,12 @@ { "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": 0.35, + "cost_per_1m_out": 0.75, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.15, + "cost_per_1m_out_cached": 0.35, "context_window": 131072, - "default_max_tokens": 65536, + "default_max_tokens": 20480, "can_reason": true, "reasoning_levels": [ "low", @@ -2266,12 +2212,12 @@ { "id": "openai/gpt-oss-20b", "name": "OpenAI: gpt-oss-20b", - "cost_per_1m_in": 0.075, - "cost_per_1m_out": 0.3, + "cost_per_1m_in": 0.05, + "cost_per_1m_out": 0.2, "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0.0375, + "cost_per_1m_out_cached": 0.05, "context_window": 131072, - "default_max_tokens": 32768, + "default_max_tokens": 65536, "can_reason": true, "reasoning_levels": [ "low", @@ -2620,12 +2566,12 @@ { "id": "qwen/qwen3-14b", "name": "Qwen: Qwen3 14B", - "cost_per_1m_in": 0.2275, - "cost_per_1m_out": 0.91, + "cost_per_1m_in": 0.12, + "cost_per_1m_out": 0.24, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, - "context_window": 131072, - "default_max_tokens": 4096, + "context_window": 40960, + "default_max_tokens": 20480, "can_reason": true, "reasoning_levels": [ "low", @@ -2686,12 +2632,12 @@ { "id": "qwen/qwen3-30b-a3b", "name": "Qwen: Qwen3 30B A3B", - "cost_per_1m_in": 0.13, - "cost_per_1m_out": 0.52, + "cost_per_1m_in": 0.08, + "cost_per_1m_out": 0.28, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, - "context_window": 131072, - "default_max_tokens": 4096, + "context_window": 40960, + "default_max_tokens": 20480, "can_reason": true, "reasoning_levels": [ "low", @@ -2770,12 +2716,12 @@ { "id": "qwen/qwen3-coder-30b-a3b-instruct", "name": "Qwen: Qwen3 Coder 30B A3B Instruct", - "cost_per_1m_in": 0.1, - "cost_per_1m_out": 0.3, + "cost_per_1m_in": 0.07, + "cost_per_1m_out": 0.27, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0, - "context_window": 262144, - "default_max_tokens": 26214, + "context_window": 160000, + "default_max_tokens": 16384, "can_reason": false, "supports_attachments": false }, @@ -3019,9 +2965,9 @@ "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": 26214, "can_reason": true, "reasoning_levels": [ "low", @@ -3205,6 +3151,18 @@ "default_reasoning_effort": "medium", "supports_attachments": false }, + { + "id": "thedrummer/rocinante-12b", + "name": "TheDrummer: Rocinante 12B", + "cost_per_1m_in": 0.17, + "cost_per_1m_out": 0.43, + "cost_per_1m_in_cached": 0, + "cost_per_1m_out_cached": 0, + "context_window": 32768, + "default_max_tokens": 16384, + "can_reason": false, + "supports_attachments": false + }, { "id": "thedrummer/unslopnemo-12b", "name": "TheDrummer: UnslopNemo 12B", @@ -3340,12 +3298,12 @@ { "id": "z-ai/glm-4.5-air", "name": "Z.ai: GLM 4.5 Air", - "cost_per_1m_in": 0.14, - "cost_per_1m_out": 0.86, + "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": 65536, + "default_max_tokens": 49152, "can_reason": true, "reasoning_levels": [ "low", @@ -3417,7 +3375,7 @@ "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, "reasoning_levels": [ "low", @@ -3448,12 +3406,12 @@ { "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": 20275, "can_reason": true, "reasoning_levels": [ "low", @@ -3466,11 +3424,11 @@ { "id": "z-ai/glm-5", "name": "Z.ai: GLM 5", - "cost_per_1m_in": 1, - "cost_per_1m_out": 3.2, + "cost_per_1m_in": 0.95, + "cost_per_1m_out": 2.55, "cost_per_1m_in_cached": 0, "cost_per_1m_out_cached": 0.2, - "context_window": 202800, + "context_window": 204800, "default_max_tokens": 65536, "can_reason": true, "reasoning_levels": [ diff --git a/internal/providers/configs/synthetic.json b/internal/providers/configs/synthetic.json index c4688124a536dc4256bcf4609481cdb3e8463f02..d5196284444574fef5e9ebe217b5e2c1f1bdcf62 100644 --- a/internal/providers/configs/synthetic.json +++ b/internal/providers/configs/synthetic.json @@ -55,6 +55,24 @@ "default_reasoning_effort": "medium", "supports_attachments": false }, + { + "id": "hf:zai-org/GLM-4.7", + "name": "GLM 4.7", + "cost_per_1m_in": 0.45, + "cost_per_1m_out": 2.19, + "cost_per_1m_in_cached": 0.45, + "cost_per_1m_out_cached": 0.45, + "context_window": 202752, + "default_max_tokens": 20275, + "can_reason": true, + "reasoning_levels": [ + "low", + "medium", + "high" + ], + "default_reasoning_effort": "medium", + "supports_attachments": false + }, { "id": "hf:zai-org/GLM-4.7-Flash", "name": "GLM 4.7 Flash", diff --git a/internal/providers/configs/venice.json b/internal/providers/configs/venice.json index c6db5f474d32359b1728a503e37b8b3933847a84..bc39bb2fb483be4a84d2068ff28fdfce2a04e644 100644 --- a/internal/providers/configs/venice.json +++ b/internal/providers/configs/venice.json @@ -529,24 +529,6 @@ "default_reasoning_effort": "medium", "supports_attachments": false }, - { - "id": "minimax-m21", - "name": "MiniMax M2.1", - "cost_per_1m_in": 0.35, - "cost_per_1m_out": 1.5, - "cost_per_1m_in_cached": 0, - "cost_per_1m_out_cached": 0, - "context_window": 198000, - "default_max_tokens": 32768, - "can_reason": true, - "reasoning_levels": [ - "low", - "medium", - "high" - ], - "default_reasoning_effort": "medium", - "supports_attachments": false - }, { "id": "minimax-m25", "name": "MiniMax M2.5",