{ "product": "oss", "outputs": { "config/litellm.yaml": { "providers": { "anthropic": { "api_key": "os.environ/ANTHROPIC_API_KEY" }, "bedrock": {}, "openai": { "api_key": "os.environ/OPENAI_API_KEY" }, "voyage": { "api_key": "os.environ/VOYAGE_API_KEY" }, "gemini": { "api_key": "os.environ/GEMINI_API_KEY" }, "minimax": { "api_key": "os.environ/MINIMAX_API_KEY" }, "deepseek": { "api_key": "os.environ/DEEPSEEK_API_KEY" }, "xai": { "api_key": "os.environ/XAI_API_KEY" }, "mistral": { "api_key": "os.environ/MISTRAL_API_KEY" }, "openrouter": { "api_key": "os.environ/OPENROUTER_API_KEY" }, "cerebras": { "api_key": "os.environ/CEREBRAS_API_KEY" }, "nvidia_nim": { "api_key": "os.environ/NVIDIA_API_KEY" }, "opencode": { "api_key": "os.environ/OPENCODE_API_KEY" }, "vercel": { "api_key": "os.environ/VERCEL_AI_GATEWAY_API_KEY" }, "hf": { "api_key": "os.environ/HF_TOKEN" }, "venice": { "api_key": "os.environ/VENICE_API_KEY" }, "nanogpt": { "api_key": "os.environ/NANOGPT_API_KEY" }, "synthetic": { "api_key": "os.environ/SYNTHETIC_API_KEY" }, "zenmux": { "api_key": "os.environ/ZENMUX_API_KEY" }, "qianfan": { "api_key": "os.environ/QIANFAN_API_KEY" }, "cfgateway": { "api_key": "os.environ/CLOUDFLARE_AI_GATEWAY_API_KEY" }, "kimi": { "api_key": "os.environ/KIMI_API_KEY" }, "cheaperinference": { "api_key": "os.environ/CHEAPER_INFERENCE_API_KEY" } }, "exclude": [ "anthropic/claude-opus-5", "anthropic/claude-sonnet-5", "openai/gpt-5.4-mini", "openai/gpt-5.6-sol", "openai/gpt-5.6-terra", "openai/gpt-5.6-luna", "gemini/gemini-3.7-flash", "gemini/gemini-3.6-flash", "gemini/gemini-3.5-flash-lite", "minimax/MiniMax-M2.7", "minimax/MiniMax-M2.5", "minimax/MiniMax-M2.5-lightning", "deepseek/deepseek-flash", "xai/grok-4.5", "mistral/mistral-medium-3-5", "mistral/mistral-large-2512", "mistral/mistral-small-2603", "openrouter/*", "openrouter/anthropic/claude-opus-5.5", "openrouter/anthropic/claude-sonnet-5.5", "openrouter/openai/gpt-6.1-sol", "openrouter/google/gemini-3.8-flash", "openrouter/deepseek/deepseek-v4.1-flash", "openrouter/anthropic/claude-opus-5", "openrouter/anthropic/claude-sonnet-5", "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b", "nvidia_nim/nvidia/nemotron-3-super-120b-a12b", "nvidia_nim/nvidia/nemotron-3-nano-30b-a3b", "nvidia_nim/nvidia/nemotron-3.5-lightning-30b-a3b", "moonshot/kimi-k3", "moonshot/kimi-k2.7-code", "moonshot/kimi-k2.6", "zai/glm-5.3", "zai/glm-5.3-flash", "zai/glm-5.3-flashx", "zai/glm-5.2", "zai/glm-5-turbo", "zai/glm-4.7-flashx", "groq/openai/gpt-oss-20b", "groq/openai/gpt-oss-120b", "cerebras/qwen-3.8-27b", "cerebras/gpt-oss-120b", "together_ai/MiniMaxAI/MiniMax-M3", "together_ai/zai-org/GLM-5.3", "together_ai/moonshotai/Kimi-K3", "fireworks_ai/accounts/fireworks/models/deepseek-v4p1-flash", "fireworks_ai/accounts/fireworks/models/glm-5p3", "fireworks_ai/accounts/fireworks/models/deepseek-v4-flash-0731", "deepinfra/deepseek-ai/DeepSeek-V4.1-Flash", "deepinfra/zai-org/GLM-5.3-Flash", "deepinfra/deepseek-ai/DeepSeek-V4-Flash-0731", "sambanova/MiniMax-M2.7", "dashscope/qwen3.8-max", "dashscope/qwen3.8-flash", "dashscope/qwen3.7-plus", "dashscope/qwen-plus", "upstage/solar-pro4", "cohere/command-a-plus-05-2026", "huggingface/openai/gpt-oss-120b", "nebius/Qwen/Qwen3-235B-A22B-Instruct-2507", "novita/zai-org/glm-5.3-flash", "siliconflow/deepseek-ai/DeepSeek-V4.1-Flash", "siliconflow/deepseek-ai/DeepSeek-V3", "requesty/anthropic/claude-sonnet-5", "vercel/anthropic/claude-opus-5", "vercel/openai/gpt-5.4", "vercel/google/gemini-3.7-flash", "vercel/moonshotai/kimi-k3", "kilo/anthropic/claude-sonnet-5", "kilo/anthropic/claude-opus-5", "kilo/anthropic/claude-haiku-4.5", "kilo/openai/gpt-5.4", "kilo/google/gemini-3.7-flash", "kilo/moonshotai/kimi-k3", "opencode_zen/kimi-k3", "opencode_zen/minimax-m3", "opencode_zen/glm-5.3", "opencode_zen/glm-5.3-flash", "opencode_zen/deepseek-v4-pro", "opencode_zen/deepseek-v4-flash", "kimi_code/k3", "kimi_code/k3-256k", "kimi_code/kimi-for-coding", "kimi_code/kimi-for-coding-highspeed", "minimax_plan/MiniMax-M3", "minimax_plan/MiniMax-M2.7", "vercel/anthropic/claude-sonnet-5.5", "vercel/openai/gpt-6.1-sol", "vercel/anthropic/claude-sonnet-5", "chutes/moonshotai/Kimi-K3-TEE", "chutes/zai-org/GLM-5.2-TEE", "chutes/deepseek-ai/DeepSeek-V4-Flash-0731-TEE", "synthetic/hf:moonshotai/Kimi-K3", "synthetic/hf:deepseek-ai/DeepSeek-V4.1-Flash", "synthetic/hf:zai-org/GLM-5.3-Flash", "ollama_cloud/kimi-k3", "ollama_cloud/glm-5.3", "ollama_cloud/deepseek-v4.1-flash", "ollama_cloud/minimax-m3", "openrouter/openai/gpt-5.6-sol", "openrouter/openai/gpt-5.6-terra", "openrouter/openai/gpt-5.6-luna", "openrouter/google/gemini-3.7-flash", "openrouter/google/gemini-3.6-flash", "openrouter/google/gemini-3.5-flash-lite", "openrouter/moonshotai/kimi-k3", "openrouter/moonshotai/kimi-k2.7-code", "openrouter/moonshotai/kimi-k2.6", "openrouter/x-ai/grok-4.5", "openrouter/x-ai/grok-4.3", "openrouter/minimax/minimax-m2.7", "openrouter/mistralai/mistral-medium-3-5", "openrouter/mistralai/mistral-large-2512", "openrouter/mistralai/mistral-small-2603", "openrouter/z-ai/glm-5.2", "openrouter/z-ai/glm-5-turbo" ], "routes": [ { "id": "anthropic/claude-fable-5-1", "patch": {} }, { "id": "anthropic/claude-opus-5-5", "patch": {} }, { "id": "anthropic/claude-sonnet-5-5", "patch": {} }, { "id": "anthropic/claude-opus-4-8", "patch": {} }, { "id": "anthropic/claude-opus-4-7", "patch": {} }, { "id": "anthropic/claude-sonnet-4-6", "patch": {} }, { "id": "anthropic/claude-haiku-4-5", "patch": {} }, { "id": "bedrock/anthropic.claude-opus-4-7", "patch": {} }, { "id": "bedrock/anthropic.claude-sonnet-4-6", "patch": {} }, { "id": "bedrock/anthropic.claude-haiku-4-5-20251001-v1:0", "patch": {} }, { "id": "openai/gpt-6-astra", "patch": {} }, { "id": "openai/gpt-6.1-sol", "patch": {} }, { "id": "openai/gpt-6-sol", "patch": {} }, { "id": "openai/gpt-6-luna", "patch": {} }, { "id": "openai/gpt-5.5", "patch": {} }, { "id": "openai/gpt-5.4", "patch": {} }, { "id": "openai/gpt-5-nano", "patch": {} }, { "id": "openai/gpt-5.3-codex", "patch": {} }, { "id": "openai/text-embedding-3-small", "patch": {} }, { "id": "voyage/voyage-3", "patch": {} }, { "id": "gemini/gemini-embedding-001", "patch": {} }, { "id": "gemini/gemini-3.8-flash", "patch": {} }, { "id": "gemini/gemini-2.5-pro", "patch": {} }, { "id": "gemini/gemini-2.5-flash", "patch": {} }, { "id": "gemini/gemini-2.5-flash-lite", "patch": {} }, { "id": "gemini/gemini-3.5-flash", "patch": {} }, { "id": "gemini/gemini-3.1-flash-lite", "patch": {} }, { "id": "minimax/MiniMax-M3", "patch": {} }, { "id": "minimax/MiniMax-M2.7-highspeed", "patch": {} }, { "private": { "model_name": "auth/claude-opus-4-8", "litellm_params": { "model": "auth/claude-opus-4-8", "additional_drop_params": [ "temperature" ], "max_tokens": 128000, "allowed_openai_params": [ "reasoning_effort" ] }, "model_info": { "input_cost_per_token": 5e-06, "output_cost_per_token": 2.5e-05 } } }, { "private": { "model_name": "auth/claude-opus-4-7", "litellm_params": { "model": "auth/claude-opus-4-7", "additional_drop_params": [ "temperature" ], "max_tokens": 128000, "allowed_openai_params": [ "reasoning_effort" ] }, "model_info": { "input_cost_per_token": 5e-06, "output_cost_per_token": 2.5e-05 } } }, { "private": { "model_name": "auth/claude-sonnet-4-6", "litellm_params": { "model": "auth/claude-sonnet-4-6", "max_tokens": 64000, "allowed_openai_params": [ "reasoning_effort" ] }, "model_info": { "input_cost_per_token": 3e-06, "output_cost_per_token": 1.5e-05 } } }, { "private": { "model_name": "auth/claude-haiku-4-5", "litellm_params": { "model": "auth/claude-haiku-4-5", "max_tokens": 64000, "allowed_openai_params": [ "reasoning_effort" ] }, "model_info": { "input_cost_per_token": 1e-06, "output_cost_per_token": 5e-06 } } }, { "private": { "model_name": "auth/claude-opus-5", "litellm_params": { "model": "auth/claude-opus-5", "additional_drop_params": [ "temperature" ], "allowed_openai_params": [ "reasoning_effort" ] }, "model_info": { "input_cost_per_token": 5e-06, "output_cost_per_token": 2.5e-05, "cache_read_input_token_cost": 5e-07, "cache_creation_input_token_cost": 6.25e-06, "supports_prompt_caching": true } } }, { "private": { "model_name": "auth/claude-sonnet-5", "litellm_params": { "model": "auth/claude-sonnet-5", "allowed_openai_params": [ "reasoning_effort" ] }, "model_info": { "input_cost_per_token": 2e-06, "output_cost_per_token": 1e-05, "cache_read_input_token_cost": 2e-07, "cache_creation_input_token_cost": 2.5e-06, "supports_prompt_caching": true } } }, { "id": "deepseek/deepseek-v4-pro", "patch": {} }, { "id": "deepseek/deepseek-v4-flash", "patch": {} }, { "id": "deepseek/deepseek-chat", "patch": {} }, { "id": "deepseek/deepseek-reasoner", "patch": {} }, { "id": "xai/grok-4.7", "patch": {} }, { "id": "xai/grok-4.3", "patch": {} }, { "id": "xai/grok-4-1-fast-reasoning", "patch": {} }, { "id": "mistral/mistral-large-latest", "patch": {} }, { "id": "mistral/codestral-latest", "patch": {} }, { "id": "openrouter/anthropic/claude-opus-4-8", "patch": {} }, { "id": "openrouter/anthropic/claude-opus-4-7", "patch": {} }, { "id": "openrouter/anthropic/claude-sonnet-4-6", "patch": {} }, { "id": "openrouter/anthropic/claude-haiku-4-5", "patch": {} }, { "id": "cerebras/llama3.1-8b", "patch": {} }, { "id": "openai/mimo-vl", "patch": { "litellm_params": { "api_key": "os.environ/XIAOMI_MIMO_API_KEY" } } }, { "id": "openai/mimo-rl", "patch": { "litellm_params": { "api_key": "os.environ/XIAOMI_MIMO_API_KEY" } } }, { "id": "openai/mimo-7b", "patch": { "litellm_params": { "api_key": "os.environ/XIAOMI_MIMO_API_KEY" } } }, { "id": "nvidia_nim/meta/llama-3.3-70b-instruct", "patch": {} }, { "id": "nvidia_nim/nvidia/llama-3.1-nemotron-70b-instruct", "patch": {} }, { "id": "nvidia_nim/meta/llama-3.2-3b-instruct", "patch": {} }, { "id": "opencode/claude-opus-4-6", "patch": {} }, { "id": "opencode/gpt-5.4", "patch": {} }, { "id": "opencode/glm-5-free", "patch": {} }, { "id": "vercel/anthropic/claude-opus-4.6", "patch": {} }, { "id": "vercel/anthropic/claude-sonnet-4.6", "patch": {} }, { "id": "vercel/anthropic/claude-haiku-4.5", "patch": {} }, { "id": "hf/deepseek-ai/DeepSeek-V3.1", "patch": {} }, { "id": "hf/meta-llama/Llama-3.3-70B-Instruct-Turbo", "patch": {} }, { "id": "hf/openai/gpt-oss-120b", "patch": {} }, { "id": "venice/claude-opus-4-6", "patch": {} }, { "id": "venice/claude-sonnet-4-6", "patch": {} }, { "id": "venice/deepseek-v4-flash", "patch": {} }, { "id": "nanogpt/anthropic/claude-opus-4.6", "patch": {} }, { "id": "nanogpt/anthropic/claude-sonnet-4.6", "patch": {} }, { "id": "nanogpt/anthropic/claude-haiku-4.5", "patch": {} }, { "id": "synthetic/hf:deepseek-ai/DeepSeek-V3.2", "patch": {} }, { "id": "synthetic/hf:meta-llama/Llama-3.3-70B-Instruct", "patch": {} }, { "id": "synthetic/hf:openai/gpt-oss-120b", "patch": {} }, { "id": "zenmux/anthropic/claude-opus-4.6", "patch": {} }, { "id": "zenmux/anthropic/claude-sonnet-4.6", "patch": {} }, { "id": "zenmux/anthropic/claude-haiku-4.5", "patch": {} }, { "id": "qianfan/ernie-4.5-turbo-128k", "patch": {} }, { "id": "qianfan/ernie-4.5-turbo-32k", "patch": {} }, { "id": "qianfan/ernie-speed-pro-128k", "patch": {} }, { "id": "cfgateway/anthropic/claude-opus-4-6", "patch": {} }, { "id": "cfgateway/anthropic/claude-sonnet-4-6", "patch": {} }, { "id": "cfgateway/anthropic/claude-haiku-4-5", "patch": {} }, { "id": "kimi/k3", "patch": {} }, { "id": "kimi/kimi-for-coding", "patch": {} }, { "id": "kimi/kimi-for-coding-highspeed", "patch": {} }, { "id": "cheaperinference/claude-sonnet-5", "patch": {} }, { "id": "cheaperinference/gpt-5.4", "patch": {} }, { "id": "cheaperinference/gpt-5.4-mini", "patch": {} } ], "settings": { "litellm_settings": { "num_retries": 3, "retry_after": 30 }, "router_settings": { "fallbacks": [ { "anthropic/claude-opus-4-8": [ "anthropic/claude-opus-4-7" ] }, { "anthropic/claude-opus-4-7": [ "anthropic/claude-sonnet-4-6" ] }, { "anthropic/claude-sonnet-4-6": [ "anthropic/claude-haiku-4-5" ] }, { "auth/claude-opus-4-8": [ "auth/claude-opus-4-7" ] }, { "auth/claude-opus-4-7": [ "auth/claude-sonnet-4-6" ] }, { "auth/claude-sonnet-4-6": [ "auth/claude-haiku-4-5" ] }, { "auth/claude-haiku-4-5": [ "anthropic/claude-haiku-4-5" ] }, { "openrouter/anthropic/claude-opus-4-8": [ "openrouter/anthropic/claude-opus-4-7" ] }, { "openrouter/anthropic/claude-opus-4-7": [ "openrouter/anthropic/claude-sonnet-4-6" ] }, { "openrouter/anthropic/claude-sonnet-4-6": [ "openrouter/anthropic/claude-haiku-4-5" ] }, { "openai/gpt-5.5": [ "openai/gpt-5.4" ] }, { "openai/gpt-5.4": [ "openai/gpt-5-nano" ] }, { "gemini/gemini-2.5-pro": [ "gemini/gemini-2.5-flash" ] }, { "gemini/gemini-2.5-flash": [ "gemini/gemini-2.5-flash-lite" ] }, { "gemini/gemini-3.5-flash": [ "gemini/gemini-3.1-flash-lite" ] }, { "minimax/MiniMax-M3": [ "minimax/MiniMax-M2.7-highspeed" ] }, { "deepseek/deepseek-v4-pro": [ "deepseek/deepseek-v4-flash" ] }, { "deepseek/deepseek-v4-flash": [ "deepseek/deepseek-chat" ] }, { "deepseek/deepseek-chat": [ "deepseek/deepseek-reasoner" ] }, { "xai/grok-4.3": [ "xai/grok-4-1-fast-reasoning" ] }, { "mistral/mistral-large-latest": [ "mistral/codestral-latest" ] }, { "nvidia_nim/meta/llama-3.3-70b-instruct": [ "nvidia_nim/nvidia/llama-3.1-nemotron-70b-instruct" ] }, { "nvidia_nim/nvidia/llama-3.1-nemotron-70b-instruct": [ "nvidia_nim/meta/llama-3.2-3b-instruct" ] }, { "openai/mimo-vl": [ "openai/mimo-rl" ] }, { "openai/mimo-rl": [ "openai/mimo-7b" ] }, { "opencode/claude-opus-4-6": [ "opencode/gpt-5.4" ] }, { "opencode/gpt-5.4": [ "opencode/glm-5-free" ] }, { "vercel/anthropic/claude-opus-4.6": [ "vercel/anthropic/claude-sonnet-4.6" ] }, { "vercel/anthropic/claude-sonnet-4.6": [ "vercel/anthropic/claude-haiku-4.5" ] }, { "hf/deepseek-ai/DeepSeek-V3.1": [ "hf/meta-llama/Llama-3.3-70B-Instruct-Turbo" ] }, { "hf/meta-llama/Llama-3.3-70B-Instruct-Turbo": [ "hf/openai/gpt-oss-120b" ] }, { "venice/claude-opus-4-6": [ "venice/claude-sonnet-4-6" ] }, { "venice/claude-sonnet-4-6": [ "venice/deepseek-v4-flash" ] }, { "nanogpt/anthropic/claude-opus-4.6": [ "nanogpt/anthropic/claude-sonnet-4.6" ] }, { "nanogpt/anthropic/claude-sonnet-4.6": [ "nanogpt/anthropic/claude-haiku-4.5" ] }, { "synthetic/hf:deepseek-ai/DeepSeek-V3.2": [ "synthetic/hf:meta-llama/Llama-3.3-70B-Instruct" ] }, { "synthetic/hf:meta-llama/Llama-3.3-70B-Instruct": [ "synthetic/hf:openai/gpt-oss-120b" ] }, { "zenmux/anthropic/claude-opus-4.6": [ "zenmux/anthropic/claude-sonnet-4.6" ] }, { "zenmux/anthropic/claude-sonnet-4.6": [ "zenmux/anthropic/claude-haiku-4.5" ] }, { "qianfan/ernie-4.5-turbo-128k": [ "qianfan/ernie-4.5-turbo-32k" ] }, { "qianfan/ernie-4.5-turbo-32k": [ "qianfan/ernie-speed-pro-128k" ] }, { "cfgateway/anthropic/claude-opus-4-6": [ "cfgateway/anthropic/claude-sonnet-4-6" ] }, { "cfgateway/anthropic/claude-sonnet-4-6": [ "cfgateway/anthropic/claude-haiku-4-5" ] } ], "routing_strategy": "simple-shuffle", "num_retries": 5, "timeout": 300, "retry_after": 30, "allowed_fails": 3, "cooldown_time": 60 }, "general_settings": { "master_key": "os.environ/LITELLM_MASTER_KEY", "database_url": "os.environ/DATABASE_URL", "store_model_in_db": false } }, "format": "yaml" }, "config/litellm_prod_catalog.json": { "providers": { "anthropic": { "api_key": "os.environ/ANTHROPIC_API_KEY" }, "openai": { "api_key": "os.environ/OPENAI_API_KEY" }, "gemini": { "api_key": "os.environ/GEMINI_API_KEY" }, "minimax": { "api_key": "os.environ/MINIMAX_API_KEY" }, "deepseek": { "api_key": "os.environ/DEEPSEEK_API_KEY" }, "xai": { "api_key": "os.environ/XAI_API_KEY" }, "mistral": { "api_key": "os.environ/MISTRAL_API_KEY" }, "openrouter": { "api_key": "os.environ/OPENROUTER_API_KEY" }, "nvidia_nim": { "api_key": "os.environ/NVIDIA_API_KEY" }, "moonshot": { "api_key": "os.environ/MOONSHOT_API_KEY" }, "zai": { "api_key": "os.environ/ZAI_API_KEY" }, "groq": { "api_key": "os.environ/GROQ_API_KEY" }, "cerebras": { "api_key": "os.environ/CEREBRAS_API_KEY" }, "together_ai": { "api_key": "os.environ/TOGETHER_API_KEY" }, "fireworks_ai": { "api_key": "os.environ/FIREWORKS_API_KEY" }, "deepinfra": { "api_key": "os.environ/DEEPINFRA_API_KEY" }, "sambanova": { "api_key": "os.environ/SAMBANOVA_API_KEY" }, "dashscope": { "api_key": "os.environ/DASHSCOPE_API_KEY" }, "upstage": { "api_key": "os.environ/UPSTAGE_API_KEY" }, "cohere": { "api_key": "os.environ/COHERE_API_KEY" }, "huggingface": { "api_key": "os.environ/HF_TOKEN" }, "nebius": { "api_key": "os.environ/NEBIUS_API_KEY" }, "novita": { "api_key": "os.environ/NOVITA_API_KEY" }, "siliconflow": { "api_key": "os.environ/SILICONFLOW_API_KEY" }, "requesty": { "api_key": "os.environ/REQUESTY_API_KEY" }, "vercel": { "api_key": "os.environ/VERCEL_AI_GATEWAY_API_KEY" }, "kilo": { "api_key": "os.environ/KILO_API_KEY" }, "opencode_zen": { "api_key": "os.environ/OPENCODE_API_KEY" }, "kimi_code": { "api_key": "os.environ/KIMI_API_KEY" }, "minimax_plan": { "api_key": "os.environ/MINIMAX_API_KEY" }, "chutes": { "api_key": "os.environ/CHUTES_API_KEY" }, "synthetic": { "api_key": "os.environ/SYNTHETIC_API_KEY" }, "ollama_cloud": { "api_key": "os.environ/OLLAMA_CLOUD_API_KEY" } }, "exclude": [ "anthropic/claude-fable-5-1", "anthropic/claude-opus-5-5", "anthropic/claude-sonnet-5-5", "anthropic/claude-opus-4-8", "anthropic/claude-opus-4-7", "anthropic/claude-sonnet-4-6", "anthropic/claude-haiku-4-5", "bedrock/anthropic.claude-opus-4-7", "bedrock/anthropic.claude-sonnet-4-6", "bedrock/anthropic.claude-haiku-4-5-20251001-v1:0", "openai/gpt-6-astra", "openai/gpt-6.1-sol", "openai/gpt-6-sol", "openai/gpt-6-luna", "openai/gpt-5.5", "openai/gpt-5.4", "openai/gpt-5-nano", "openai/gpt-5.3-codex", "openai/text-embedding-3-small", "voyage/voyage-3", "gemini/gemini-embedding-001", "gemini/gemini-3.8-flash", "gemini/gemini-2.5-pro", "gemini/gemini-2.5-flash", "gemini/gemini-2.5-flash-lite", "gemini/gemini-3.5-flash", "gemini/gemini-3.1-flash-lite", "minimax/MiniMax-M3", "minimax/MiniMax-M2.7-highspeed", "deepseek/deepseek-v4-pro", "deepseek/deepseek-v4-flash", "deepseek/deepseek-chat", "deepseek/deepseek-reasoner", "xai/grok-4.7", "xai/grok-4.3", "xai/grok-4-1-fast-reasoning", "mistral/mistral-large-latest", "mistral/codestral-latest", "openrouter/anthropic/claude-opus-4-8", "openrouter/anthropic/claude-opus-4-7", "openrouter/anthropic/claude-sonnet-4-6", "openrouter/anthropic/claude-haiku-4-5", "cerebras/llama3.1-8b", "openai/mimo-vl", "openai/mimo-rl", "openai/mimo-7b", "nvidia_nim/meta/llama-3.3-70b-instruct", "nvidia_nim/nvidia/llama-3.1-nemotron-70b-instruct", "nvidia_nim/meta/llama-3.2-3b-instruct", "opencode/claude-opus-4-6", "opencode/gpt-5.4", "opencode/glm-5-free", "vercel/anthropic/claude-opus-4.6", "vercel/anthropic/claude-sonnet-4.6", "vercel/anthropic/claude-haiku-4.5", "hf/deepseek-ai/DeepSeek-V3.1", "hf/meta-llama/Llama-3.3-70B-Instruct-Turbo", "hf/openai/gpt-oss-120b", "venice/claude-opus-4-6", "venice/claude-sonnet-4-6", "venice/deepseek-v4-flash", "nanogpt/anthropic/claude-opus-4.6", "nanogpt/anthropic/claude-sonnet-4.6", "nanogpt/anthropic/claude-haiku-4.5", "synthetic/hf:deepseek-ai/DeepSeek-V3.2", "synthetic/hf:meta-llama/Llama-3.3-70B-Instruct", "synthetic/hf:openai/gpt-oss-120b", "zenmux/anthropic/claude-opus-4.6", "zenmux/anthropic/claude-sonnet-4.6", "zenmux/anthropic/claude-haiku-4.5", "qianfan/ernie-4.5-turbo-128k", "qianfan/ernie-4.5-turbo-32k", "qianfan/ernie-speed-pro-128k", "cfgateway/anthropic/claude-opus-4-6", "cfgateway/anthropic/claude-sonnet-4-6", "cfgateway/anthropic/claude-haiku-4-5", "kimi/k3", "kimi/kimi-for-coding", "kimi/kimi-for-coding-highspeed", "cheaperinference/claude-sonnet-5", "cheaperinference/gpt-5.4", "cheaperinference/gpt-5.4-mini", "openrouter/openai/gpt-5.6-sol", "openrouter/openai/gpt-5.6-terra", "openrouter/openai/gpt-5.6-luna", "openrouter/google/gemini-3.7-flash", "openrouter/google/gemini-3.6-flash", "openrouter/google/gemini-3.5-flash-lite", "openrouter/moonshotai/kimi-k3", "openrouter/moonshotai/kimi-k2.7-code", "openrouter/moonshotai/kimi-k2.6", "openrouter/x-ai/grok-4.5", "openrouter/x-ai/grok-4.3", "openrouter/minimax/minimax-m2.7", "openrouter/mistralai/mistral-medium-3-5", "openrouter/mistralai/mistral-large-2512", "openrouter/mistralai/mistral-small-2603", "openrouter/z-ai/glm-5.2", "openrouter/z-ai/glm-5-turbo" ], "routes": [ { "id": "anthropic/claude-opus-5", "patch": {} }, { "id": "anthropic/claude-sonnet-5", "patch": {} }, { "id": "openai/gpt-5.4-mini", "patch": {} }, { "id": "openai/gpt-5.6-sol", "patch": {} }, { "id": "openai/gpt-5.6-terra", "patch": {} }, { "id": "openai/gpt-5.6-luna", "patch": {} }, { "id": "gemini/gemini-3.7-flash", "patch": {} }, { "id": "gemini/gemini-3.6-flash", "patch": {} }, { "id": "gemini/gemini-3.5-flash-lite", "patch": {} }, { "id": "minimax/MiniMax-M2.7", "patch": {} }, { "id": "minimax/MiniMax-M2.5", "patch": {} }, { "id": "minimax/MiniMax-M2.5-lightning", "patch": {} }, { "id": "deepseek/deepseek-flash", "patch": {} }, { "id": "xai/grok-4.5", "patch": {} }, { "id": "mistral/mistral-medium-3-5", "patch": {} }, { "id": "mistral/mistral-large-2512", "patch": {} }, { "id": "mistral/mistral-small-2603", "patch": {} }, { "id": "openrouter/*", "patch": {} }, { "id": "openrouter/anthropic/claude-opus-5.5", "patch": {} }, { "id": "openrouter/anthropic/claude-sonnet-5.5", "patch": {} }, { "id": "openrouter/openai/gpt-6.1-sol", "patch": {} }, { "id": "openrouter/google/gemini-3.8-flash", "patch": {} }, { "id": "openrouter/deepseek/deepseek-v4.1-flash", "patch": {} }, { "id": "openrouter/anthropic/claude-opus-5", "patch": {} }, { "id": "openrouter/anthropic/claude-sonnet-5", "patch": {} }, { "id": "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b", "patch": {} }, { "id": "nvidia_nim/nvidia/nemotron-3-super-120b-a12b", "patch": {} }, { "id": "nvidia_nim/nvidia/nemotron-3-nano-30b-a3b", "patch": {} }, { "id": "nvidia_nim/nvidia/nemotron-3.5-lightning-30b-a3b", "patch": {} }, { "id": "moonshot/kimi-k3", "patch": {} }, { "id": "moonshot/kimi-k2.7-code", "patch": {} }, { "id": "moonshot/kimi-k2.6", "patch": {} }, { "id": "zai/glm-5.3", "patch": {} }, { "id": "zai/glm-5.3-flash", "patch": {} }, { "id": "zai/glm-5.3-flashx", "patch": {} }, { "id": "zai/glm-5.2", "patch": {} }, { "id": "zai/glm-5-turbo", "patch": {} }, { "id": "zai/glm-4.7-flashx", "patch": {} }, { "private": { "model_name": "runyour/gemini-3.5-flash", "litellm_params": { "model": "openai/gemini/gemini-3.5-flash", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 1.5e-06, "output_cost_per_token": "9e-06", "cache_read_input_token_cost": 1.5e-06, "cache_creation_input_token_cost": 1.5e-06 } } }, { "private": { "model_name": "runyour/kimi-k3", "litellm_params": { "model": "openai/moonshot/kimi-k3", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY", "additional_drop_params": [ "temperature" ] }, "model_info": { "input_cost_per_token": "3e-06", "output_cost_per_token": 1.5e-05, "cache_read_input_token_cost": "3e-06", "cache_creation_input_token_cost": "3e-06" } } }, { "private": { "model_name": "runyour/deepseek-v4-pro", "litellm_params": { "model": "openai/deepseek/deepseek-v4-pro", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 1.32e-06, "output_cost_per_token": 3.96e-06, "cache_read_input_token_cost": 1.32e-06, "cache_creation_input_token_cost": 1.32e-06 } } }, { "private": { "model_name": "runyour/qwen3.7-max", "litellm_params": { "model": "openai/qwen/qwen3.7-max", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 2.5e-06, "output_cost_per_token": 7.5e-06, "cache_read_input_token_cost": 2.5e-06, "cache_creation_input_token_cost": 2.5e-06 } } }, { "private": { "model_name": "runyour/grok-4.5", "litellm_params": { "model": "openai/xai/grok-4.5", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": "2e-06", "output_cost_per_token": "6e-06", "cache_read_input_token_cost": "2e-06", "cache_creation_input_token_cost": "2e-06" } } }, { "private": { "model_name": "runyour/claude-fable-5", "litellm_params": { "model": "openai/anthropic/claude-fable-5", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": "1e-05", "output_cost_per_token": "5e-05", "cache_read_input_token_cost": "1e-05", "cache_creation_input_token_cost": "1e-05" } } }, { "private": { "model_name": "runyour/gpt-5.4-nano", "litellm_params": { "model": "openai/openai/gpt-5.4-nano", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": "2e-07", "output_cost_per_token": 1.2533333333333333e-06, "cache_read_input_token_cost": "2e-07", "cache_creation_input_token_cost": "2e-07" } } }, { "private": { "model_name": "runyour/deepseek-v4-flash", "litellm_params": { "model": "openai/deepseek/deepseek-v4-flash", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 4.4e-07, "output_cost_per_token": 1.32e-06, "cache_read_input_token_cost": 4.4e-07, "cache_creation_input_token_cost": 4.4e-07 } } }, { "private": { "model_name": "runyour/qwen3.7-plus", "litellm_params": { "model": "openai/qwen/qwen3.7-plus", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": "4e-07", "output_cost_per_token": 1.6e-06, "cache_read_input_token_cost": "4e-07", "cache_creation_input_token_cost": "4e-07" } } }, { "private": { "model_name": "runyour/grok-4-20", "litellm_params": { "model": "openai/xai/grok-4.20-reasoning", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 1.25e-06, "output_cost_per_token": 2.5e-06, "cache_read_input_token_cost": 1.25e-06, "cache_creation_input_token_cost": 1.25e-06 } } }, { "private": { "model_name": "runyour/solar-pro3", "litellm_params": { "model": "openai/upstage/solar-pro3", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 1.5e-07, "output_cost_per_token": "6e-7", "cache_read_input_token_cost": 1.5e-07, "cache_creation_input_token_cost": 1.5e-07 } } }, { "id": "groq/openai/gpt-oss-20b", "patch": {} }, { "id": "groq/openai/gpt-oss-120b", "patch": {} }, { "id": "cerebras/qwen-3.8-27b", "patch": {} }, { "id": "cerebras/gpt-oss-120b", "patch": {} }, { "id": "together_ai/MiniMaxAI/MiniMax-M3", "patch": {} }, { "id": "together_ai/zai-org/GLM-5.3", "patch": {} }, { "id": "together_ai/moonshotai/Kimi-K3", "patch": {} }, { "id": "fireworks_ai/accounts/fireworks/models/deepseek-v4p1-flash", "patch": {} }, { "id": "fireworks_ai/accounts/fireworks/models/glm-5p3", "patch": {} }, { "id": "fireworks_ai/accounts/fireworks/models/deepseek-v4-flash-0731", "patch": {} }, { "id": "deepinfra/deepseek-ai/DeepSeek-V4.1-Flash", "patch": {} }, { "id": "deepinfra/zai-org/GLM-5.3-Flash", "patch": {} }, { "id": "deepinfra/deepseek-ai/DeepSeek-V4-Flash-0731", "patch": {} }, { "id": "sambanova/MiniMax-M2.7", "patch": {} }, { "id": "dashscope/qwen3.8-max", "patch": {} }, { "id": "dashscope/qwen3.8-flash", "patch": {} }, { "id": "dashscope/qwen3.7-plus", "patch": {} }, { "id": "dashscope/qwen-plus", "patch": {} }, { "id": "upstage/solar-pro4", "patch": {} }, { "id": "cohere/command-a-plus-05-2026", "patch": {} }, { "id": "huggingface/openai/gpt-oss-120b", "patch": {} }, { "id": "nebius/Qwen/Qwen3-235B-A22B-Instruct-2507", "patch": {} }, { "id": "novita/zai-org/glm-5.3-flash", "patch": {} }, { "id": "siliconflow/deepseek-ai/DeepSeek-V4.1-Flash", "patch": {} }, { "id": "siliconflow/deepseek-ai/DeepSeek-V3", "patch": {} }, { "id": "requesty/anthropic/claude-sonnet-5", "patch": {} }, { "id": "vercel/anthropic/claude-opus-5", "patch": {} }, { "id": "vercel/openai/gpt-5.4", "patch": {} }, { "id": "vercel/google/gemini-3.7-flash", "patch": {} }, { "id": "vercel/moonshotai/kimi-k3", "patch": {} }, { "id": "kilo/anthropic/claude-sonnet-5", "patch": {} }, { "id": "kilo/anthropic/claude-opus-5", "patch": {} }, { "id": "kilo/anthropic/claude-haiku-4.5", "patch": {} }, { "id": "kilo/openai/gpt-5.4", "patch": {} }, { "id": "kilo/google/gemini-3.7-flash", "patch": {} }, { "id": "kilo/moonshotai/kimi-k3", "patch": {} }, { "id": "opencode_zen/kimi-k3", "patch": {} }, { "id": "opencode_zen/minimax-m3", "patch": {} }, { "id": "opencode_zen/glm-5.3", "patch": {} }, { "id": "opencode_zen/glm-5.3-flash", "patch": {} }, { "id": "opencode_zen/deepseek-v4-pro", "patch": {} }, { "id": "opencode_zen/deepseek-v4-flash", "patch": {} }, { "id": "kimi_code/k3", "patch": {} }, { "id": "kimi_code/k3-256k", "patch": {} }, { "id": "kimi_code/kimi-for-coding", "patch": {} }, { "id": "kimi_code/kimi-for-coding-highspeed", "patch": {} }, { "id": "minimax_plan/MiniMax-M3", "patch": {} }, { "id": "minimax_plan/MiniMax-M2.7", "patch": {} }, { "id": "vercel/anthropic/claude-sonnet-5.5", "patch": {} }, { "id": "vercel/openai/gpt-6.1-sol", "patch": {} }, { "id": "vercel/anthropic/claude-sonnet-5", "patch": {} }, { "private": { "model_name": "runyour/gpt-5.5", "litellm_params": { "model": "openai/openai/gpt-5.5", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": "5e-06", "output_cost_per_token": "3e-05", "cache_read_input_token_cost": "5e-06", "cache_creation_input_token_cost": "5e-06" } } }, { "private": { "model_name": "runyour/gpt-5.4", "litellm_params": { "model": "openai/openai/gpt-5.4", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 2.5e-06, "output_cost_per_token": 1.5e-05, "cache_read_input_token_cost": 2.5e-06, "cache_creation_input_token_cost": 2.5e-06 } } }, { "private": { "model_name": "runyour/gpt-5.4-mini", "litellm_params": { "model": "openai/openai/gpt-5.4-mini", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 7.5e-07, "output_cost_per_token": 4.5e-06, "cache_read_input_token_cost": 7.5e-07, "cache_creation_input_token_cost": 7.5e-07 } } }, { "private": { "model_name": "runyour/claude-opus-5", "litellm_params": { "model": "openai/anthropic/claude-opus-5", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": "5e-06", "output_cost_per_token": 2.5e-05, "cache_read_input_token_cost": "5e-06", "cache_creation_input_token_cost": "5e-06" } } }, { "private": { "model_name": "runyour/claude-sonnet-5", "litellm_params": { "model": "openai/anthropic/claude-sonnet-5", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": "2e-06", "output_cost_per_token": "1e-05", "cache_read_input_token_cost": "2e-06", "cache_creation_input_token_cost": "2e-06" } } }, { "private": { "model_name": "runyour/claude-haiku-4-5", "litellm_params": { "model": "openai/anthropic/claude-haiku-4-5", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": "1e-06", "output_cost_per_token": "5e-06", "cache_read_input_token_cost": "1e-06", "cache_creation_input_token_cost": "1e-06" } } }, { "private": { "model_name": "runyour/gemini-3.7-flash", "litellm_params": { "model": "openai/gemini/gemini-3.7-flash", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 7.077533333329999e-07, "output_cost_per_token": 3.5387533333330003e-06, "cache_read_input_token_cost": 7.077533333329999e-07, "cache_creation_input_token_cost": 7.077533333329999e-07 } } }, { "private": { "model_name": "runyour/gemini-3.6-flash", "litellm_params": { "model": "openai/gemini/gemini-3.6-flash", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 6.964e-07, "output_cost_per_token": 3.4820000000000003e-06, "cache_read_input_token_cost": 6.964e-07, "cache_creation_input_token_cost": 6.964e-07 } } }, { "private": { "model_name": "runyour/kimi-k2.7-code", "litellm_params": { "model": "openai/moonshot/kimi-k2.7-code", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 9.499999999999999e-07, "output_cost_per_token": "4e-06", "cache_read_input_token_cost": 9.499999999999999e-07, "cache_creation_input_token_cost": 9.499999999999999e-07 } } }, { "private": { "model_name": "runyour/qwen3.8-max", "litellm_params": { "model": "openai/qwen/qwen3.8-max", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": "2e-06", "output_cost_per_token": "6e-06", "cache_read_input_token_cost": "2e-06", "cache_creation_input_token_cost": "2e-06" } } }, { "private": { "model_name": "runyour/qwen3.7-flash", "litellm_params": { "model": "openai/qwen/qwen3.7-flash", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 2.0000000000000002e-07, "output_cost_per_token": 8.000000000000001e-07, "cache_read_input_token_cost": 2.0000000000000002e-07, "cache_creation_input_token_cost": 2.0000000000000002e-07 } } }, { "private": { "model_name": "runyour/grok-4.3", "litellm_params": { "model": "openai/xai/grok-4.3", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 1.25e-06, "output_cost_per_token": 2.5e-06, "cache_read_input_token_cost": 1.25e-06, "cache_creation_input_token_cost": 1.25e-06 } } }, { "private": { "model_name": "runyour/solar-pro4", "litellm_params": { "model": "openai/upstage/solar-pro4", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": "3e-8", "output_cost_per_token": 1.2e-07, "cache_read_input_token_cost": "3e-8", "cache_creation_input_token_cost": "3e-8" } } }, { "private": { "model_name": "runyour/glm-5-turbo", "litellm_params": { "model": "openai/zai/glm-5.2", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 1.4e-06, "output_cost_per_token": 4.4e-06, "cache_read_input_token_cost": 1.4e-06, "cache_creation_input_token_cost": 1.4e-06 } } }, { "private": { "model_name": "runyour/glm-5.2", "litellm_params": { "model": "openai/zai/glm-5.2", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 1.4e-06, "output_cost_per_token": 4.4e-06, "cache_read_input_token_cost": 1.4e-06, "cache_creation_input_token_cost": 1.4e-06 } } }, { "private": { "model_name": "runyour/minimax-m3", "litellm_params": { "model": "openai/minimax/minimax-m3", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": "3e-7", "output_cost_per_token": 1.2e-06, "cache_read_input_token_cost": "3e-7", "cache_creation_input_token_cost": "3e-7" } } }, { "private": { "model_name": "runyour/minimax-m2.7", "litellm_params": { "model": "openai/minimax/minimax-m2.7", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": "3e-7", "output_cost_per_token": 1.2e-06, "cache_read_input_token_cost": "3e-7", "cache_creation_input_token_cost": "3e-7" } } }, { "private": { "model_name": "runyour/mimo-v2.5-pro", "litellm_params": { "model": "openai/xiaomi/mimo-v2.5-pro", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 4.35e-07, "output_cost_per_token": 8.7e-07, "cache_read_input_token_cost": 4.35e-07, "cache_creation_input_token_cost": 4.35e-07 } } }, { "private": { "model_name": "runyour/mimo-v2.5", "litellm_params": { "model": "openai/xiaomi/mimo-v2.5", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 1.4e-07, "output_cost_per_token": 2.8e-07, "cache_read_input_token_cost": 1.4e-07, "cache_creation_input_token_cost": 1.4e-07 } } }, { "private": { "model_name": "runyour/glm-5.3", "litellm_params": { "model": "openai/zai/glm-5.3", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 1.4e-06, "output_cost_per_token": 4.4e-06, "cache_read_input_token_cost": 1.4e-06, "cache_creation_input_token_cost": 1.4e-06 } } }, { "private": { "model_name": "runyour/qwen3.8-flash", "litellm_params": { "model": "openai/qwen/qwen3.8-flash", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 1.5e-07, "output_cost_per_token": 4.6999999999999995e-07, "cache_read_input_token_cost": 1.5e-07, "cache_creation_input_token_cost": 1.5e-07 } } }, { "private": { "model_name": "runyour/nemotron-3-ultra-550b-a55b", "litellm_params": { "model": "openai/nvidia/nemotron-3-ultra-550b-a55b", "api_base": "os.environ/RUNYOUR_API_BASE", "api_key": "os.environ/RUNYOUR_API_KEY" }, "model_info": { "input_cost_per_token": 0, "output_cost_per_token": 0, "cache_read_input_token_cost": 0, "cache_creation_input_token_cost": 0 } } }, { "id": "chutes/moonshotai/Kimi-K3-TEE", "patch": {} }, { "id": "chutes/zai-org/GLM-5.2-TEE", "patch": {} }, { "id": "chutes/deepseek-ai/DeepSeek-V4-Flash-0731-TEE", "patch": {} }, { "id": "synthetic/hf:moonshotai/Kimi-K3", "patch": {} }, { "id": "synthetic/hf:deepseek-ai/DeepSeek-V4.1-Flash", "patch": {} }, { "id": "synthetic/hf:zai-org/GLM-5.3-Flash", "patch": {} }, { "id": "ollama_cloud/kimi-k3", "patch": {} }, { "id": "ollama_cloud/glm-5.3", "patch": {} }, { "id": "ollama_cloud/deepseek-v4.1-flash", "patch": {} }, { "id": "ollama_cloud/minimax-m3", "patch": {} } ], "settings": {}, "format": "optional" } }, "cli": { "path": "clients/cli/src/commands/currentProdModels.ts", "groups": { "Current prod / anthropic": [ "anthropic/claude-fable-5-1", "anthropic/claude-opus-5-5", "anthropic/claude-opus-5", "anthropic/claude-sonnet-5-5", "anthropic/claude-sonnet-5", "anthropic/claude-haiku-4-5" ], "Current prod / auth": [ "auth/claude-opus-5", "auth/claude-opus-4-8", "auth/claude-opus-4-7", "auth/claude-sonnet-4-6", "auth/claude-sonnet-5", "auth/claude-haiku-4-5", "auth/gpt-6-astra", "auth/gpt-6.1-sol", "auth/gpt-6-luna", "auth/gpt-5.6-sol", "auth/gpt-5.6-terra", "auth/gpt-5.6-luna" ], "Current prod / cerebras": [ "cerebras/qwen-3.8-27b", "cerebras/gpt-oss-120b" ], "Current prod / chutes": [ "chutes/moonshotai/Kimi-K3-TEE", "chutes/zai-org/GLM-5.2-TEE", "chutes/deepseek-ai/DeepSeek-V4-Flash-0731-TEE" ], "Current prod / cohere": [ "cohere/command-a-plus-05-2026" ], "Current prod / dashscope": [ "dashscope/qwen3.8-max", "dashscope/qwen3.8-flash", "dashscope/qwen3.7-plus", "dashscope/qwen-plus" ], "Current prod / deepinfra": [ "deepinfra/deepseek-ai/DeepSeek-V4.1-Flash", "deepinfra/zai-org/GLM-5.3-Flash", "deepinfra/deepseek-ai/DeepSeek-V4-Flash-0731" ], "Current prod / deepseek": [ "deepseek/deepseek-flash", "deepseek/deepseek-v4-pro", "deepseek/deepseek-v4-flash", "deepseek/deepseek-chat", "deepseek/deepseek-reasoner" ], "Current prod / fireworks_ai": [ "fireworks_ai/accounts/fireworks/models/deepseek-v4p1-flash", "fireworks_ai/accounts/fireworks/models/glm-5p3", "fireworks_ai/accounts/fireworks/models/deepseek-v4-flash-0731" ], "Current prod / gemini": [ "gemini/gemini-3.8-flash", "gemini/gemini-3.7-flash", "gemini/gemini-3.6-flash", "gemini/gemini-3.5-flash", "gemini/gemini-3.5-flash-lite", "gemini/gemini-2.5-pro", "gemini/gemini-2.5-flash", "gemini/gemini-2.5-flash-lite" ], "Current prod / groq": [ "groq/openai/gpt-oss-20b", "groq/openai/gpt-oss-120b" ], "Current prod / huggingface": [ "huggingface/openai/gpt-oss-120b" ], "Current prod / kilo": [ "kilo/anthropic/claude-sonnet-5", "kilo/anthropic/claude-opus-5", "kilo/anthropic/claude-haiku-4.5", "kilo/openai/gpt-5.4", "kilo/google/gemini-3.7-flash", "kilo/moonshotai/kimi-k3" ], "Current prod / kimi_code": [ "kimi_code/k3", "kimi_code/k3-256k", "kimi_code/kimi-for-coding", "kimi_code/kimi-for-coding-highspeed" ], "Current prod / minimax": [ "minimax/MiniMax-M3", "minimax/MiniMax-M2.7", "minimax/MiniMax-M2.7-highspeed", "minimax/MiniMax-M2.5", "minimax/MiniMax-M2.5-lightning" ], "Current prod / minimax_plan": [ "minimax_plan/MiniMax-M3", "minimax_plan/MiniMax-M2.7" ], "Current prod / mistral": [ "mistral/mistral-medium-3-5", "mistral/mistral-large-2512", "mistral/mistral-small-2603", "mistral/mistral-large-latest", "mistral/codestral-latest" ], "Current prod / moonshot": [ "moonshot/kimi-k3", "moonshot/kimi-k2.7-code", "moonshot/kimi-k2.6" ], "Current prod / nebius": [ "nebius/Qwen/Qwen3-235B-A22B-Instruct-2507" ], "Current prod / novita": [ "novita/zai-org/glm-5.3-flash" ], "Current prod / nvidia_nim": [ "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b", "nvidia_nim/nvidia/nemotron-3-super-120b-a12b", "nvidia_nim/nvidia/nemotron-3-nano-30b-a3b", "nvidia_nim/nvidia/nemotron-3.5-lightning-30b-a3b", "nvidia_nim/meta/llama-3.3-70b-instruct", "nvidia_nim/nvidia/llama-3.1-nemotron-70b-instruct", "nvidia_nim/meta/llama-3.2-3b-instruct" ], "Current prod / ollama_cloud": [ "ollama_cloud/kimi-k3", "ollama_cloud/glm-5.3", "ollama_cloud/deepseek-v4.1-flash", "ollama_cloud/minimax-m3" ], "Current prod / openai": [ "openai/gpt-6-astra", "openai/gpt-6.1-sol", "openai/gpt-6-sol", "openai/gpt-6-luna", "openai/gpt-5.4", "openai/gpt-5.4-mini", "openai/gpt-5.6-sol", "openai/gpt-5.6-terra", "openai/gpt-5.6-luna" ], "Current prod / opencode_zen": [ "opencode_zen/kimi-k3", "opencode_zen/minimax-m3", "opencode_zen/glm-5.3", "opencode_zen/glm-5.3-flash", "opencode_zen/deepseek-v4-pro", "opencode_zen/deepseek-v4-flash" ], "Current prod / openrouter": [ "openrouter/anthropic/claude-opus-5.5", "openrouter/anthropic/claude-sonnet-5.5", "openrouter/openai/gpt-6.1-sol", "openrouter/google/gemini-3.8-flash", "openrouter/deepseek/deepseek-v4.1-flash", "openrouter/anthropic/claude-opus-5", "openrouter/anthropic/claude-sonnet-5", "openrouter/anthropic/claude-opus-4-8", "openrouter/anthropic/claude-opus-4-7", "openrouter/anthropic/claude-sonnet-4-6", "openrouter/anthropic/claude-haiku-4-5" ], "Current prod / requesty": [ "requesty/anthropic/claude-sonnet-5" ], "Current prod / runyour": [ "runyour/gemini-3.5-flash", "runyour/kimi-k3", "runyour/deepseek-v4-pro", "runyour/qwen3.7-max", "runyour/grok-4.5", "runyour/claude-fable-5", "runyour/gpt-5.4-nano", "runyour/deepseek-v4-flash", "runyour/qwen3.7-plus", "runyour/grok-4-20", "runyour/solar-pro3", "runyour/gpt-5.5", "runyour/gpt-5.4", "runyour/gpt-5.4-mini", "runyour/claude-opus-5", "runyour/claude-sonnet-5", "runyour/claude-haiku-4-5", "runyour/gemini-3.7-flash", "runyour/gemini-3.6-flash", "runyour/kimi-k2.7-code", "runyour/qwen3.8-max", "runyour/qwen3.7-flash", "runyour/grok-4.3", "runyour/solar-pro4", "runyour/glm-5-turbo", "runyour/glm-5.2", "runyour/minimax-m3", "runyour/minimax-m2.7", "runyour/mimo-v2.5-pro", "runyour/mimo-v2.5", "runyour/glm-5.3", "runyour/qwen3.8-flash", "runyour/nemotron-3-ultra-550b-a55b" ], "Current prod / sambanova": [ "sambanova/MiniMax-M2.7" ], "Current prod / siliconflow": [ "siliconflow/deepseek-ai/DeepSeek-V4.1-Flash", "siliconflow/deepseek-ai/DeepSeek-V3" ], "Current prod / synthetic": [ "synthetic/hf:moonshotai/Kimi-K3", "synthetic/hf:deepseek-ai/DeepSeek-V4.1-Flash", "synthetic/hf:zai-org/GLM-5.3-Flash" ], "Current prod / together_ai": [ "together_ai/MiniMaxAI/MiniMax-M3", "together_ai/zai-org/GLM-5.3", "together_ai/moonshotai/Kimi-K3" ], "Current prod / upstage": [ "upstage/solar-pro4" ], "Current prod / vercel": [ "vercel/anthropic/claude-opus-5", "vercel/anthropic/claude-haiku-4.5", "vercel/openai/gpt-5.4", "vercel/google/gemini-3.7-flash", "vercel/moonshotai/kimi-k3", "vercel/anthropic/claude-sonnet-5.5", "vercel/openai/gpt-6.1-sol", "vercel/anthropic/claude-sonnet-5" ], "Current prod / xai": [ "xai/grok-4.7", "xai/grok-4.5", "xai/grok-4.3", "xai/grok-4-1-fast-reasoning" ], "Current prod / zai": [ "zai/glm-5.3", "zai/glm-5.3-flash", "zai/glm-5.3-flashx", "zai/glm-5.2", "zai/glm-5-turbo", "zai/glm-4.7-flashx" ] } } }