nRouter for AI Crawlers

nRouter is a managed LLM gateway: one OpenAI-compatible API key reaches every model in the catalogue, with routing, failover, guardrails, budgets and exact cost tracking handled by the gateway. This page is a plain, text-only index of the site for AI crawlers, agents and low-bandwidth access. It lists the models in the live catalogue by modality, the provider clouds behind them, and the product, use-case and documentation pages, each with a short description and a canonical link.

For a shorter machine-readable overview see /llms.txt. For the full text see /llms-full.txt. For the complete list of URLs see /sitemap.xml. Every model page also has a plain-text twin at /models/{model}/llms.txt.

Site

• Models — Every model served through the gateway, with specs and exact list pricing.
• Leaderboard — Models plotted on a price-versus-performance frontier from live benchmark data.
• Pricing — Pay as you go. Models are served at exact provider list price with a 0% platform fee.
• Documentation — API reference, SDKs, frameworks and guides for building on the gateway.
• Blog — Engineering deep dives, guides and company posts from the nRouter team.
• Help Center — The fastest path to an answer: channels, quick answers and support tiers.
• Support — Open a support ticket for account, billing and API issues.
• Enterprise — Dedicated tenancy, security review support and custom terms for large teams.
• Book a demo — A live walkthrough of the gateway with the team.
• Status — Per-component health for the gateway, dashboard and providers, with incident history.

Chat models

• Amazon Nova Lite V1 — Amazon Nova Lite V1 is a chat model made by AWS Bedrock, served on nRouter. It has a 300K-token context window.
• Amazon Nova Micro V1 — Amazon Nova Micro V1 is a chat model made by AWS Bedrock, served on nRouter. It has a 128K-token context window.
• Amazon Nova Pro V1 — Amazon Nova Pro V1 is a chat model made by AWS Bedrock, served on nRouter. It has a 300K-token context window.
• Chat Latest — Chat Latest is a chat model made by OpenAI, served on nRouter.
• Claude Fable 5 — Claude Fable 5 is a multimodal foundation model supporting text and image inputs with a 1,000,000-token context window and 128,000 max output tokens. It features a January 2026 knowledge cutoff for agentic workflows.
• Claude Fable 5 1 — Claude Fable 5 1 is a chat model made by Anthropic, served on nRouter. It features a dense transformer architecture for complex multi-turn instruction, long-context retrieval, and code generation. It has a 1M-token context window.
• Claude Haiku 4 5 20251001 — Claude Haiku 4.5 from Anthropic — 200K-token context, multimodal vision, function calling.
• Claude Opus 4 5 20251101 — Claude Opus 4.5 from Anthropic — 200K-token context, multimodal vision, function calling.
• Claude Opus 4 6 — Claude Opus 4.6 is a multimodal coding and reasoning model with a 1,000,000-token context window and a 128,000 max output token limit. It processes text and image inputs for large codebases and complex agent operations.
• Claude Opus 4 7 — Claude Opus 4.7 is a deep reasoning model with an upgraded tokenizer and high-resolution vision, engineered for long-horizon software engineering and interface design.
• Claude Opus 4 8 — Claude Opus 4.8 is an enterprise-grade model built for sustained collaboration, dynamic workflow execution, and large-scale architectural problem solving.
• Claude Opus 5 — Claude Opus 5 is a multimodal foundation model supporting text and image inputs with a 1,000,000-token context window and 128,000 max output tokens. It has a May 2026 knowledge cutoff for multi-stage software engineering.
• Claude Opus 5 5 — Claude Opus 5 is a chat model made by Anthropic, served on nRouter. It features a dense transformer architecture for complex multi-turn instruction, long-context retrieval, and code generation. It has a 1M-token context window.
• Claude Sonnet 4 5 20250929 — Claude Sonnet 4.5 from Anthropic — 200K-token context, multimodal vision, function calling.
• Claude Sonnet 4 6 — Claude Sonnet 4.6 provides balanced speed and intelligence with a 1M token context window, designed for enterprise software engineering and agentic planning.
• Claude Sonnet 5 — Claude Sonnet 5 is a multimodal foundation model offering a 1,000,000-token context window and 128,000 max output tokens. It supports text and image inputs with a January 2026 knowledge cutoff for production workflows.
• Claude Sonnet 5 5 — Claude Sonnet 5 is a chat model made by Anthropic, served on nRouter. It features a dense transformer architecture for complex multi-turn instruction, long-context retrieval, and code generation. It has a 1M-token context window.
• Deepseek R1 0528 Maas — Deepseek R1 0528 Maas is a chat model made by DeepSeek, served on nRouter. It is designed for deliberative reasoning with autonomous planning and mathematical logic decomposition.
• Deepseek V3.2 Maas — Deepseek V3.2 Maas is a chat model made by DeepSeek, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Deepseek V4 Flash 0731 — DeepSeek-V4-Flash-0731 is the production release of DeepSeek-V4-Flash, incorporating an integrated speculative decoding module. It delivers enhanced agentic performance, tool execution, and code synthesis across a 1M-token context window.
• Deepseek V4 Pro — DeepSeek-V4-Pro is a text model with a 1,048,576-token context window, 393,216 max output tokens, evaluated at 87.5 on MMLU-Pro, 90.1 on GPQA Diamond, and 80.6 on SWE-bench Verified.
• Deepseek V4 Pro 0813 — DeepSeek-V4-Pro-0813 is the production release of the DeepSeek-V4-Pro architecture, pairing 1.6T MoE parameters with a DSpark speculative decoding engine for accelerated agentic execution, full-stack software engineering, and deep logical reasoning.
• Deepseek V4.1 Flash — Deepseek V4.1 Flash is a chat model made by DeepSeek, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Devstral 2 123b — Devstral 2 123b is a chat model made by Mistral AI, served on nRouter. It features a dense transformer architecture for structured instruction following, reasoning, and code synthesis.
• Gemini 3 Flash Preview — Gemini 3 Flash Preview is a chat model made by Google, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Gemini 3.1 Flash Lite — Gemini 3.1 Flash-Lite is Google's lightweight multimodal foundation model for high-throughput tasks. It processes up to 1,048,576 input tokens across text, audio, video, image, and PDF, with 65,536 max output tokens for translation, transcription, and classification.
• Gemini 3.1 Pro Preview — Gemini 3.1 Pro Preview is a chat model made by Google, served on nRouter.
• Gemini 3.5 Flash — Gemini 3.5 Flash is a chat model made by Google, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction. It has a 1M-token context window.
• Gemini 3.5 Flash Lite — Gemini 3.5 Flash Lite is a chat model made by Google, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction. It has a 1M-token context window.
• Gemini 3.5 Live Translate Preview — Gemini 3.5 Live Translate Preview is a chat model made by Google, served on nRouter.
• Gemini 3.6 Flash — Gemini 3.6 Flash is a chat model made by Google, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction. It has a 1M-token context window.
• Gemini 3.7 Flash — Gemini 3.7 Flash is a chat model made by Google, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction. It has a 1M-token context window.
• Gemini 3.8 Flash — Gemini 3.8 Flash is a chat model made by Google, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction. It has a 1M-token context window.
• Gemini Live 2.5 Flash Native Audio — Gemini Live 2.5 Flash Native Audio is a chat model made by Google, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Gemma 3 12b It — Gemma 3 12b It is a chat model made by Google, served on nRouter.
• Gemma 3 27b It — Gemma 3 27b It is a chat model made by Google, served on nRouter.
• Gemma 3 4b It — Gemma 3 4b It is a chat model made by Google, served on nRouter.
• Glm 4.7 — Glm 4.7 is a chat model made by Z-AI, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Glm 4.7 Flash — Glm 4.7 Flash is a chat model made by Z-AI, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Glm 4.7 Maas — Glm 4.7 Maas is a chat model made by Z-AI, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Glm 5 — Glm 5 is a chat model made by Z-AI, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Glm 5 Maas — Glm 5 Maas is a chat model made by Z-AI, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Glm 5.1 — Glm 5.1 is a chat model made by Z-AI, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation. It has a 200K-token context window.
• Glm 5.2 — Glm 5.2 is a chat model made by Z-AI, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation. It has a 1M-token context window.
• Glm 5.2 Fast Preview — Glm 5.2 Fast Preview is a chat model made by Z-AI, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Glm 5.2 Maas — Glm 5.2 Maas is a chat model made by Z-AI, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Glm 5.3 — Glm 5.3 is a chat model made by Z-AI, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• GLM 5.3 — GLM 5.3 is a chat model made by Z-AI, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Glm 5.3 Prime — Glm 5.3 Prime is a chat model made by Z-AI, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Global.Amazon.Nova 2 Lite V1:0 — Global.Amazon.Nova 2 Lite V1:0 is a chat model made by AWS Bedrock, served on nRouter. It has a 1M-token context window.
• Gpt 3.5 Turbo — OpenAI's legacy compact language model optimized for conversational interactions and general text completion tasks.
• Gpt 3.5 Turbo 0125 — Updated snapshot of GPT-3.5 Turbo with improved instruction following, higher accuracy at responding in requested formats, and bug fixes.
• Gpt 3.5 Turbo 1106 — Snapshot of GPT-3.5 Turbo supporting 16K context window, improved instruction following, JSON mode, and parallel function calling.
• Gpt 3.5 Turbo 16k — Legacy 16K context variant of GPT-3.5 Turbo offering expanded token capacity for longer text processing tasks.
• Gpt 3.5 Turbo Instruct — Gpt 3.5 Turbo Instruct is a chat model made by OpenAI, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Gpt 4 — GPT-4 from OpenAI — 8K-token context, multimodal vision, function calling.
• Gpt 4 0613 — Snapshot of GPT-4 introduced in June 2023 with function calling support and improved steering capabilities.
• Gpt 4 Turbo — GPT-4 Turbo from OpenAI — 128K-token context, multimodal vision, function calling.
• Gpt 4 Turbo 2024 04 09 — Generally available GPT-4 Turbo with Vision snapshot supporting structured JSON, parallel tool calling, and multimodal inputs.
• Gpt 4.1 — GPT-4.1 from OpenAI — 1M-token context, multimodal vision, function calling.
• Gpt 4.1 2025 04 14 — Primary snapshot of GPT-4.1 offering million-token context processing, multimodal inputs, and high-volume reasoning support.
• Gpt 4.1 Mini — GPT-4.1 Mini from OpenAI — 1M-token context, multimodal vision, function calling.
• Gpt 4.1 Mini 2025 04 14 — Snapshot of GPT-4.1 Mini providing 1,047,576 context tokens and high-speed execution for lightweight reasoning and data parsing.
• Gpt 4.1 Nano — GPT-4.1 Nano from OpenAI — 1M-token context, multimodal vision, function calling.
• Gpt 4.1 Nano 2025 04 14 — Dated release of GPT-4.1 Nano combining sub-dollar token economics with large-context ingestion and rapid classification capability.
• Gpt 4o — GPT-4o from OpenAI — 128K-token context, multimodal vision, function calling.
• Gpt 4o 2024 05 13 — Original launch snapshot of GPT-4o offering multimodal text and vision input capabilities with 4,096 max output tokens.
• Gpt 4o 2024 08 06 — Enhanced GPT-4o snapshot introducing Structured Outputs with 100% schema reliability and increased 16,384 token output limits.
• Gpt 4o 2024 11 20 — Refined GPT-4o checkpoint with improved creative writing naturalness, structured responses, and file analysis capabilities.
• Gpt 4o Mini — GPT-4o Mini from OpenAI — 128K-token context, multimodal vision, function calling.
• Gpt 4o Mini 2024 07 18 — Launch snapshot of GPT-4o Mini optimized for low-latency chat, structured document extraction, and high-frequency API calls.
• Gpt 4o Mini Search Preview — Gpt 4o Mini Search Preview is a chat model made by OpenAI, served on nRouter. It features a vision multimodal architecture for visual document parsing, image analysis, and spatial reasoning.
• Gpt 4o Mini Search Preview 2025 03 11 — Gpt 4o Mini Search Preview 2025 03 11 is a chat model made by OpenAI, served on nRouter. It features a vision multimodal architecture for visual document parsing, image analysis, and spatial reasoning.
• Gpt 4o Search Preview — Gpt 4o Search Preview is a chat model made by OpenAI, served on nRouter. It features a vision multimodal architecture for visual document parsing, image analysis, and spatial reasoning.
• Gpt 4o Search Preview 2025 03 11 — Gpt 4o Search Preview 2025 03 11 is a chat model made by OpenAI, served on nRouter. It features a vision multimodal architecture for visual document parsing, image analysis, and spatial reasoning.
• Gpt 5 — GPT-5 from OpenAI — 272K-token context, multimodal vision, function calling.
• Gpt 5 2025 08 07 — OpenAI's August 7, 2025 production snapshot of GPT-5, providing deep reasoning capabilities across a 400,000-token context window with configurable reasoning effort for complex coding, math, and agentic workflows.
• Gpt 5 Mini — GPT-5 Mini from OpenAI — 272K-token context, multimodal vision, function calling.
• Gpt 5 Mini 2025 08 07 — Initial snapshot of GPT-5 Mini featuring 400K context capacity and up to 128K generation tokens for cost-efficient reasoning.
• Gpt 5 Nano — GPT-5 Nano from OpenAI — 272K-token context, multimodal vision, function calling.
• Gpt 5 Nano 2025 08 07 — Snapshot of GPT-5 Nano offering 400K context input support and fast inference for cost-sensitive automation pipelines.
• Gpt 5 Pro — Gpt 5 Pro is a reasoning model made by OpenAI, served on nRouter.
• Gpt 5 Pro 2025 10 06 — Gpt 5 Pro 2025 10 06 is a reasoning model made by OpenAI, served on nRouter.
• Gpt 5 Search Api — Specialized GPT-5 model variant designed for web search execution via the Chat Completions API, supporting a 200,000-token context window with automated search retrieval before generating completions.
• Gpt 5 Search Api 2025 10 14 — Gpt 5 Search Api 2025 10 14 is a reasoning model made by OpenAI, served on nRouter.
• Gpt 5.1 — GPT-5.1 from OpenAI — 272K-token context, multimodal vision, function calling.
• Gpt 5.1 2025 11 13 — Snapshot of GPT-5.1 featuring 400K context input and 128K output limits for reliable production agent deployment.
• Gpt 5.1 Chat Latest — Gpt 5.1 Chat Latest is a reasoning model made by OpenAI, served on nRouter.
• Gpt 5.2 — GPT-5.2 from OpenAI — 272K-token context, multimodal vision, function calling.
• Gpt 5.2 2025 12 11 — Pinned snapshot of GPT-5.2 released on December 11, 2025, offering deterministic reasoning behavior and 400K token context capacity for production systems.
• Gpt 5.2 Chat Latest — Gpt 5.2 Chat Latest is a reasoning model made by OpenAI, served on nRouter.
• Gpt 5.2 Pro — Gpt 5.2 Pro is a reasoning model made by OpenAI, served on nRouter.
• Gpt 5.2 Pro 2025 12 11 — Gpt 5.2 Pro 2025 12 11 is a reasoning model made by OpenAI, served on nRouter.
• Gpt 5.3 Chat Latest — Gpt 5.3 Chat Latest is a reasoning model made by OpenAI, served on nRouter.
• Gpt 5.3 Codex — OpenAI's specialized reasoning model fine-tuned for agentic software engineering and long-horizon programming tasks in Codex, featuring a 400,000-token context window and up to 128,000 output tokens.
• Gpt 5.4 — GPT-5.4 from OpenAI — 1M-token context, multimodal vision, function calling.
• Gpt 5.4 2026 03 05 — Deterministic snapshot of GPT-5.4 from March 5, 2026, featuring a 1,050,000-token context window and 128,000 max output tokens. It processes text and image inputs with an August 2025 knowledge cutoff.
• Gpt 5.4 Mini — GPT-5.4 Mini from OpenAI — 1M-token context, multimodal vision, function calling.
• Gpt 5.4 Mini 2026 03 17 — Specific snapshot of GPT-5.4 Mini from March 17, 2026, delivering balanced reasoning performance and 400K context capacity at reduced inference latency.
• Gpt 5.4 Nano — GPT-5.4 Nano from OpenAI — 1M-token context, multimodal vision, function calling.
• Gpt 5.4 Nano 2026 03 17 — March 17, 2026 snapshot of GPT-5.4 Nano offering rapid response times and 400K context processing for high-volume, budget-conscious production workloads.
• Gpt 5.4 Pro — Gpt 5.4 Pro is a reasoning model made by OpenAI, served on nRouter.
• Gpt 5.4 Pro 2026 03 05 — Gpt 5.4 Pro 2026 03 05 is a reasoning model made by OpenAI, served on nRouter.
• Gpt 5.5 — GPT-5.5 from OpenAI — 1M-token context, multimodal vision, function calling.
• Gpt 5.5 2026 04 23 — Fixed snapshot of GPT-5.5 dated April 23, 2026, providing stable reasoning outputs and 1.05 million token context limits for production pipelines.
• Gpt 5.5 Pro — Gpt 5.5 Pro is a reasoning model made by OpenAI, served on nRouter.
• Gpt 5.5 Pro 2026 04 23 — Gpt 5.5 Pro 2026 04 23 is a reasoning model made by OpenAI, served on nRouter.
• Gpt 5.6 Luna — Gpt 5.6 Luna is a reasoning model made by OpenAI, served on nRouter. It has a 1.1M-token context window.
• Gpt 5.6 Sol — Gpt 5.6 Sol is a reasoning model made by OpenAI, served on nRouter. It has a 1.1M-token context window.
• Gpt 5.6 Terra — Gpt 5.6 Terra is a reasoning model made by OpenAI, served on nRouter. It has a 1.1M-token context window.
• Gpt 6 Astra — Gpt 6 Astra is a reasoning model made by OpenAI, served on nRouter. It is designed for deliberative reasoning with autonomous planning and mathematical logic decomposition. It has a 200K-token context window.
• Gpt 6 Luna — Gpt 6 Luna is a reasoning model made by OpenAI, served on nRouter. It is designed for deliberative reasoning with autonomous planning and mathematical logic decomposition.
• Gpt 6 Sol — Gpt 6 Sol is a reasoning model made by OpenAI, served on nRouter. It is designed for deliberative reasoning with autonomous planning and mathematical logic decomposition. It has a 200K-token context window.
• Gpt 6.1 Sol — Gpt 6.1 Sol is a reasoning model made by OpenAI, served on nRouter. It is designed for deliberative reasoning with autonomous planning and mathematical logic decomposition. It has a 200K-token context window.
• Gpt Oss 120b 1 — Gpt Oss 120b 1 is a chat model made by OpenAI, served on nRouter.
• Gpt Oss 120b Maas — Gpt Oss 120b Maas is a chat model made by OpenAI, served on nRouter.
• Gpt Oss 20b 1 — Gpt Oss 20b 1 is a chat model made by OpenAI, served on nRouter.
• Gpt Oss Safeguard 120b — Gpt Oss Safeguard 120b is a chat model made by OpenAI, served on nRouter.
• Gpt Oss Safeguard 20b — Gpt Oss Safeguard 20b is a chat model made by OpenAI, served on nRouter.
• Grok 4 1 Fast Non Reasoning — Grok 4 1 Fast Non Reasoning is a reasoning model made by xAI, served on nRouter. It is designed for deliberative reasoning with autonomous planning and mathematical logic decomposition.
• Grok 4 1 Fast Reasoning — Grok 4 1 Fast Reasoning is a reasoning model made by xAI, served on nRouter. It is designed for deliberative reasoning with autonomous planning and mathematical logic decomposition.
• Jev — Jev is a chat model made by TypeSafe AI, served on nRouter. It is designed for sub-second deterministic classification and routing without generative token latency. It has a 128K-token context window.
• Kimi K2 Thinking — Kimi K2 Thinking is a reasoning model made by Moonshot AI, served on nRouter.
• Kimi K2 Thinking Maas — Kimi K2 Thinking Maas is a chat model made by Moonshot AI, served on nRouter.
• Kimi K2.5 — Kimi K2.5 is a chat model made by Moonshot AI, served on nRouter.
• Kimi K2.7 Code — Kimi K2.7 Code is a code model made by Moonshot AI, served on nRouter. It has a 256K-token context window.
• Kimi K3 — Kimi K3 is a chat model made by Moonshot AI, served on nRouter. It has a 1M-token context window.
• Kimi K3 — Kimi K3 is a chat model made by Moonshot AI, served on nRouter.
• Llama 3.3 70B Instruct — Llama 3.3 70B Instruct is a chat model made by Meta, served on nRouter. It features an open-weights dense transformer architecture for instruction following, synthesis, and code generation.
• Llama 4 Maverick 17B 128E Instruct FP8 — Llama 4 Maverick 17B 128E Instruct FP8 is a chat model made by Meta, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Llama 4 Scout 17B 16E Instruct — Llama 4 Scout 17B 16E Instruct is a chat model made by Meta, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Llama3 1 70b Instruct — Llama3 1 70b Instruct is a chat model made by Meta, served on nRouter. It features an open-weights dense transformer architecture for instruction following, synthesis, and code generation.
• Llama3 1 8b Instruct — Llama3 1 8b Instruct is a chat model made by Meta, served on nRouter. It features an open-weights dense transformer architecture for instruction following, synthesis, and code generation.
• Llama3 3 70b Instruct — Llama3 3 70b Instruct is a chat model made by Meta, served on nRouter. It features an open-weights dense transformer architecture for instruction following, synthesis, and code generation.
• Llama4 Maverick 17b Instruct — Llama4 Maverick 17b Instruct is a chat model made by Meta, served on nRouter. It features an open-weights dense transformer architecture for instruction following, synthesis, and code generation.
• Llama4 Scout 17b Instruct — Llama4 Scout 17b Instruct is a chat model made by Meta, served on nRouter. It features an open-weights dense transformer architecture for instruction following, synthesis, and code generation.
• Magistral Small 2509 — Magistral Small 2509 is a chat model made by Mistral AI, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Minimax M2 — Minimax M2 is a chat model made by AWS Bedrock, served on nRouter.
• Minimax M2 Maas — Minimax M2 Maas is a chat model made by Google Vertex AI, served on nRouter.
• Minimax M2.1 — Minimax M2.1 is a chat model made by AWS Bedrock, served on nRouter.
• Minimax M2.5 — Minimax M2.5 is a chat model made by AWS Bedrock, served on nRouter.
• Ministral 3 14b Instruct — Ministral 3 14b Instruct is a chat model made by Mistral AI, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Ministral 3 3b Instruct — Ministral 3 3b Instruct is a chat model made by Mistral AI, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Ministral 3 8b Instruct — Ministral 3 8b Instruct is a chat model made by Mistral AI, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Mistral Large 3 675b Instruct — Mistral Large 3 675b Instruct is a chat model made by Mistral AI, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Mistral Medium 2505 — Mistral Medium 2505 is a chat model made by Mistral AI, served on nRouter. It features a dense transformer architecture for structured instruction following, reasoning, and code synthesis.
• Mistral Small 2503 — Mistral Small 2503 is a chat model made by Mistral AI, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Nemotron Nano 12b V2 — Nemotron Nano 12b V2 is a chat model made by AWS Bedrock, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Nemotron Nano 3 30b — Nemotron Nano 3 30b is a chat model made by AWS Bedrock, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Nemotron Nano 9b V2 — Nemotron Nano 9b V2 is a chat model made by AWS Bedrock, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Nemotron Super 3 120b — Nemotron Super 3 120b is a chat model made by AWS Bedrock, served on nRouter.
• Nova 2 Lite — Nova 2 Lite is a chat model made by AWS Bedrock, served on nRouter.
• Nova Lite — Nova Lite is a chat model made by AWS Bedrock, served on nRouter.
• O1 — o1 from OpenAI — 200K-token context, multimodal vision, function calling.
• O1 2024 12 17 — December 17, 2024 checkpoint of OpenAI o1 with a 200K token context window and 100K output ceiling, ranking 146th on LM Arena text benchmark.
• O1 Pro — O1 Pro is a reasoning model made by OpenAI, served on nRouter. It is designed for deliberative reasoning with autonomous planning and mathematical logic decomposition.
• O1 Pro 2025 03 19 — Compute-intensive version of o1 dated March 19, 2025, using additional reasoning compute for high-difficulty analytical and mathematical queries.
• O3 — o3 from OpenAI — 200K-token context, multimodal vision, function calling.
• O3 2025 04 16 — April 16, 2025 checkpoint of OpenAI o3, setting high benchmarks on mathematics, code generation, and visual reasoning across a 200,000-token context window with up to 100,000 generation tokens.
• O3 Mini — o3-mini from OpenAI — 200K-token context, multimodal vision, function calling.
• O3 Mini 2025 01 31 — January 31, 2025 snapshot of o3-mini delivering fast reasoning capabilities, function calling, and 200K token context limits for production services.
• O4 Mini — o4-mini from OpenAI — 200K-token context, multimodal vision, function calling.
• O4 Mini 2025 04 16 — April 16, 2025 snapshot of o4-mini featuring multimodal perception and 100K token output generation, ranked 160 on the LM Arena text leaderboard.
• Palmyra Vision 7b — Palmyra Vision 7b is a chat model made by AWS Bedrock, served on nRouter. It features a vision multimodal architecture for visual document parsing, image analysis, and spatial reasoning.
• Phi 4 — Phi-4 is Microsoft's 14-billion parameter language model engineered for complex reasoning, mathematical problem solving, and code generation, trained on synthetic reasoning datasets and curated web texts.
• Phi 4 Mini Instruct — Phi-4-mini-instruct is Microsoft's 3.8-billion parameter lightweight instruction-tuned model featuring a 128K context window, engineered for high-throughput edge and latency-sensitive server workloads.
• Phi 4 Mini Reasoning — Phi-4-mini-reasoning is Microsoft's compact reasoning model delivering chain-of-thought problem solving in mathematics and logic within a 128K token context window.
• Phi 4 Multimodal Instruct — Phi-4-multimodal-instruct is Microsoft's 5.6-billion parameter multimodal model accepting text, vision, and audio inputs with a 128K context window for cross-modality comprehension.
• Phi 4 Reasoning — Phi-4-reasoning is Microsoft's 14-billion parameter reasoning model specialized in step-by-step logical deduction, STEM proofs, and rigorous code verification.
• Pixtral Large 2502 — Pixtral Large 2502 is a chat model made by Mistral AI, served on nRouter. It features a vision multimodal architecture for visual document parsing, image analysis, and spatial reasoning.
• Qwen Flash — Qwen Flash is Alibaba lightweight and cost-effective foundation model offering 1M-token context support with structured output and context caching for high-speed, high-volume production serving.
• Qwen Flash Character — Qwen Flash Character is Alibaba Cloud's lightweight role-playing dialogue model tuned for conversational agents, NPC game characters, and persona simulation with an 8K context window.
• Qwen Max — Qwen Max is an Alibaba Cloud proprietary dense text model supporting a 32,768-token context window for enterprise reasoning workloads.
• Qwen Mt Flash — Qwen MT Flash is a lightweight machine translation model designed for low-latency, high-concurrency cross-lingual translation tasks supporting 92 languages with a 16K context window.
• Qwen Mt Lite — Qwen MT Lite is a lightweight machine translation model from Alibaba Cloud optimized for high-throughput multilingual translation across 92 languages within a 16K context window.
• Qwen Mt Plus — Qwen-MT-Plus is Alibaba's enhanced machine translation model supporting 92 languages and dialects, providing terminology customization and domain adaptation for cross-lingual enterprise documentation.
• Qwen Plus — Qwen Plus provides a balance of speed and reasoning quality for general conversational and instruction tasks.
• Qwen Plus 2025 01 25 — Qwen-Plus-2025-01-25 is a dated snapshot of Alibaba Cloud Qwen-Plus model, providing balanced speed and reasoning across a 1M-token context window with support for function calling and structured outputs.
• Qwen Plus 2025 04 28 — Qwen Plus 2025-04-28 is a snapshot release of Alibaba's balanced language model offering a 1M-token context window with native function calling and structured outputs. It delivers high throughput for general conversational assistance and long-form document synthesis.
• Qwen Plus 2025 07 14 — Qwen-Plus-2025-07-14 is the July 14, 2025 dated snapshot of Alibaba's Qwen Plus foundation model, providing fixed weights and a 1M-token context window for reliable enterprise pipelines and document analysis.
• Qwen Plus 2025 12 01 — Qwen Plus 2025-12-01 is a dated production snapshot of Alibaba Cloud's balanced foundation model, offering stable execution for general instruction following, summarization, and reasoning across a 1M-token context window.
• Qwen Plus Character — Qwen Plus Character is a specialized conversational model fine-tuned for character roleplay and persona-driven interactive dialogue with a 32K context window.
• Qwen Plus Latest — Qwen Plus Latest is the rolling production pointer to Alibaba latest Qwen Plus foundation model checkpoint.
• Qwen Turbo — Qwen Turbo is a cost-effective language model engineered for low-latency dialogue and high-throughput serving.
• Qwen Vl Ocr 2025 11 20 — Qwen-VL-OCR-2025-11-20 is a specialized visual text extraction model from Alibaba Cloud for document parsing, table structure extraction, and mathematical formula recognition from images.
• Qwen2 7b Instruct — Qwen2 7b Instruct is a chat model made by Qwen, served on nRouter.
• Qwen3 235b A22b 2507 — Qwen3 235b A22b 2507 is a chat model made by Qwen, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Qwen3 235b A22b Instruct 2507 Maas — Qwen3 235b A22b Instruct 2507 Maas is a chat model made by Qwen, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Qwen3 32b — Qwen3 32B is a dense language model designed for complex coding, math, and structured tool use.
• Qwen3 Coder 30b A3b — Qwen3 Coder 30b A3b is a chat model made by Qwen, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Qwen3 Coder 480b A35b — Qwen3 Coder 480b A35b is a chat model made by Qwen, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Qwen3 Coder 480b A35b Instruct Maas — Qwen3 Coder 480b A35b Instruct Maas is a chat model made by Qwen, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Qwen3 Coder Flash — Qwen3-Coder-Flash is a lightweight, low-latency code intelligence model featuring a 1M-token context window designed for rapid code autocompletion, real-time code reviews, and high-throughput software agent workflows.
• Qwen3 Livetranslate Flash — Qwen3 Livetranslate Flash is a chat model made by Qwen, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Qwen3 Livetranslate Flash 2025 12 01 — Qwen3 Livetranslate Flash 2025 12 01 is a chat model made by Qwen, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Qwen3 Next 80b A3b — Qwen3 Next 80b A3b is a chat model made by Qwen, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Qwen3 Next 80b A3b Instruct Maas — Qwen3 Next 80b A3b Instruct Maas is a chat model made by Qwen, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Qwen3 Next 80b A3b Thinking Maas — Qwen3 Next 80b A3b Thinking Maas is a chat model made by Qwen, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Qwen3 Omni Flash — Qwen3-Omni-Flash is an omni-modal foundation model supporting real-time processing of text, image, audio, and video inputs with hybrid thinking modes and natural speech output across 11 languages.
• Qwen3 Omni Flash Realtime 2025 09 15 — Qwen3 Omni Flash Realtime 2025 09 15 is a chat model made by Qwen, served on nRouter. It features a vision multimodal architecture for visual document parsing, image analysis, and spatial reasoning.
• Qwen3 S2s Flash Realtime — Qwen3 S2s Flash Realtime is a chat model made by Qwen, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Qwen3 Vl 235b A22b — Qwen3 Vl 235b A22b is a chat model made by Qwen, served on nRouter. It features a vision multimodal architecture for visual document parsing, image analysis, and spatial reasoning.
• Qwen3 Vl Plus — Qwen3-VL-Plus is Alibaba's multimodal vision-language model supporting high-resolution image analysis and video comprehension up to 1 hour in length. It features a 256K-token context window with native function calling across visual and textual reasoning tasks.
• Qwen3.5 122b A10b — Sparse mixture-of-experts architecture activating 10 billion parameters of 122 billion total, combining gated delta networks with long-context comprehension.
• Qwen3.5 27b — Dense 27-billion parameter language model offering efficient inference and native long-context processing for general text and reasoning tasks.
• Qwen3.5 35b A3b — Sparse mixture-of-experts model activating 3 billion parameters from a 35-billion parameter base, designed for low-latency text generation and agent interactions.
• Qwen3.5 397b A17b — High-capacity mixture-of-experts foundation model with 397 billion total parameters and 17 billion active parameters, providing broad domain knowledge, math reasoning, and coding expertise.
• Qwen3.5 Flash — Qwen3.5-Flash is a high-speed, cost-efficient language model from Alibaba Cloud supporting a 1M-token context window, native thinking modes, structured outputs, and function calling for high-throughput applications.
• Qwen3.5 Flash 2026 02 23 — Qwen3.5 Flash 2026-02-23 is a lightweight foundation language model checkpoint from Alibaba engineered for fast inference and complex reasoning with a 1M-token context window.
• Qwen3.5 Livetranslate Flash Realtime — Qwen3.5 Livetranslate Flash Realtime is a chat model made by Qwen, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Qwen3.5 Livetranslate Flash Realtime 2026 05 19 — Qwen3.5 Livetranslate Flash Realtime 2026 05 19 is a chat model made by Qwen, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Qwen3.5 Omni Flash — Qwen3.5-Omni-Flash is a lightweight next-generation omni-modal model supporting text, audio, and visual inputs across 113 input languages with low-latency streaming text and speech outputs.
• Qwen3.5 Omni Flash 2026 03 15 — Qwen3.5-Omni-Flash-2026-03-15 is a dated snapshot of Alibaba's omni-modal model supporting low-latency streaming interactions across text, audio, image, and video inputs with native speech output.
• Qwen3.5 Omni Flash Realtime — Qwen3.5 Omni Flash Realtime is a chat model made by Qwen, served on nRouter. It features a vision multimodal architecture for visual document parsing, image analysis, and spatial reasoning.
• Qwen3.5 Omni Flash Realtime 2026 03 15 — Qwen3.5 Omni Flash Realtime 2026 03 15 is a chat model made by Qwen, served on nRouter. It features a vision multimodal architecture for visual document parsing, image analysis, and spatial reasoning.
• Qwen3.5 Omni Plus — Qwen3.5-Omni-Plus is an omni-modal foundation model from Alibaba Cloud supporting native speech, audio, image, and video inputs alongside text, engineered for real-time interactive multi-sensory tasks.
• Qwen3.5 Omni Plus 2026 03 15 — Qwen3.5-Omni-Plus 2026-03-15 is an omni-modal foundation model capable of processing text, audio, images, and video within a 64K-token context window. Designed for low-latency interactive processing, it supports function calling and video analysis for real-time applications.
• Qwen3.5 Omni Plus Realtime — Qwen3.5 Omni Plus Realtime is a chat model made by Qwen, served on nRouter. It features a vision multimodal architecture for visual document parsing, image analysis, and spatial reasoning.
• Qwen3.5 Omni Plus Realtime 2026 03 15 — Qwen3.5 Omni Plus Realtime 2026 03 15 is a chat model made by Qwen, served on nRouter. It features a vision multimodal architecture for visual document parsing, image analysis, and spatial reasoning.
• Qwen3.5 Plus — Qwen3.5-Plus is Alibaba Cloud's foundation multimodal model offering a 1M-token context window, 64K maximum output tokens, and native vision-language reasoning for conversational and agentic workflows.
• Qwen3.5 Plus 2026 02 15 — Qwen3.5 Plus 2026-02-15 is a multimodal snapshot model delivering native text, image, and video processing with a 1M-token context window and up to 64K output tokens for agentic workflows.
• Qwen3.5 Plus 2026 04 20 — Qwen3.5-Plus-2026-04-20 is a pinned production snapshot of Alibaba Cloud's general-purpose Qwen3.5-Plus model, supporting thinking mode, structured output generation, and tool use across a 1M-token context window.
• Qwen3.6 27b — Qwen3.6-27B is a 27-billion parameter dense multimodal language model from Alibaba featuring a vision encoder, native 262K context window, and enhanced repository-level agentic coding capabilities.
• Qwen3.6 35b A3b — Qwen3.6-35B-A3B is an open-weight mixture-of-experts multimodal model from Alibaba activating 3 billion of its 35 billion parameters, designed for efficient agentic coding with a 262K native context.
• Qwen3.6 Flash — Qwen3.6 Flash is a high-throughput language model from Alibaba featuring native tool calling, hybrid thinking modes, and a 1M-token context window for low-latency agentic workflows.
• Qwen3.6 Flash 2026 04 16 — Qwen3.6-Flash-2026-04-16 is a high-efficiency multimodal model featuring a 1M context window and 64K max output tokens, engineered for fast processing across text, image, and video modalities.
• Qwen3.6 Plus — Qwen3.6-Plus is a high-capacity multimodal model from Alibaba featuring 1M context, 64K max output, and native understanding across text, images, and video, ranked #79 on LMSYS Arena.
• Qwen3.6 Plus 2026 04 02 — Qwen3.6-Plus 2026-04-02 is a snapshot of Alibaba's multimodal foundation model featuring a 1M-token context window and 64K max output tokens. It integrates native thinking modes with multimodal inputs across text, images, and video for complex agentic workflows.
• Qwen3.7 Flash — Qwen3.7-Flash is Alibaba's high-speed multimodal model providing a 1M-token context window, 64K output token limit, thinking modes, and fast processing across text, high-resolution images, and long-form video.
• Qwen3.7 Flash 2026 07 15 — Qwen3.7 Flash 2026-07-15 is a high-throughput multimodal snapshot model designed for fast text, image, and video understanding with a 1M-token context window and 64K maximum output tokens.
• Qwen3.7 Max — Qwen3.7-Max is an Alibaba Cloud dense reasoning model featuring dual thinking and non-thinking operational modes, a 1,000,000-token context window, and agentic tool use for enterprise workflows.
• Qwen3.7 Max 2026 05 17 — Qwen3.7 Max 2026 05 17 is a chat model made by Qwen, served on nRouter.
• Qwen3.7 Max 2026 05 20 — Qwen3.7 Max 2026-05-20 is an Alibaba dense reasoning model checkpoint delivering logical deduction, structured outputs, and agentic tool execution across a 1,000,000-token context window.
• Qwen3.7 Max 2026 06 08 — Qwen3.7-Max-2026-06-08 is an Alibaba Cloud foundation model supporting hybrid thinking mode and a 1,048,576-token context window for text reasoning and enterprise problem solving.
• Qwen3.7 Max Preview — Qwen3.7 Max Preview is a chat model made by Qwen, served on nRouter.
• Qwen3.7 Plus — Qwen3.7-Plus is a balanced multimodal foundation model supporting thinking and standard modes across a 1M-token context window, with 64K output token capacity for document synthesis and agentic tool use.
• Qwen3.7 Plus 2026 05 26 — Qwen3.7-Plus-2026-05-26 is a production snapshot of Alibaba's multimodal model with a 1M-token context window and 64K max output, capable of processing text, high-resolution images, and long-form video.
• Qwen3.8 2.4t A95b — Qwen3.8 2.4t A95b is a chat model made by Qwen, served on nRouter. It features a sparse Mixture-of-Experts architecture for multi-turn instruction and code generation.
• Qwen3.8 27b — Qwen3.8-27B is Alibaba's 27-billion parameter dense foundation model featuring native vision-language understanding, flexible thinking modes, and autonomous multi-step agent execution.
• Qwen3.8 Flash — Qwen3.8-Flash is Alibaba's high-efficiency production model offering 1M context length, sparse attention architecture, and fast reasoning across text and vision modalities.
• Qwen3.8 Livetranslate Flash Realtime — Qwen3.8 Livetranslate Flash Realtime is a chat model made by Qwen, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Qwen3.8 Max — Qwen3.8-Max is an Alibaba foundation model offering a 1,000,000-token context window, hybrid thinking modes, and multimodal reasoning across text, image, and video inputs. It provides tool calling and long-context comprehension for complex tasks.
• Qwen3.8 Max 0902 — Qwen3.8 Max 0902 is a chat model made by Qwen, served on nRouter.
• Qwen3.8 Omni Flash — Qwen3.8 Omni Flash is a chat model made by Qwen, served on nRouter. It features a vision multimodal architecture for visual document parsing, image analysis, and spatial reasoning.
• Qwq Plus 2025 03 05 — Qwq Plus 2025 03 05 is a reasoning model made by Alibaba US, served on nRouter. It is designed for deliberative reasoning with autonomous planning and mathematical logic decomposition.
• R1 — R1 is a reasoning model made by AWS Bedrock, served on nRouter. It is designed for deliberative reasoning with autonomous planning and mathematical logic decomposition.
• V3 — V3 is a chat model made by AWS Bedrock, served on nRouter.
• V3.2 — V3.2 is a chat model made by AWS Bedrock, served on nRouter.
• Voxtral Mini 3b 2507 — Voxtral Mini 3b 2507 is a chat model made by AWS Bedrock, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.
• Voxtral Small 24b 2507 — Voxtral Small 24b 2507 is a chat model made by AWS Bedrock, served on nRouter. It features a compact low-latency architecture for high-throughput multi-turn chat and structured extraction.

Image models

• Chatgpt Image Latest — Image generation model.
• Gpt Image 1 — Multimodal OpenAI image synthesis model accepting text and visual prompts to generate and edit images with inpainting support.
• Gpt Image 1 Mini — Cost-efficient OpenAI image generation model designed for high-throughput visual asset creation and rapid prototyping.
• Gpt Image 1.5 — Enhanced multimodal generation model by OpenAI offering refined prompt adherence, image editing, and text-image generation outputs.
• Gpt Image 2 — Image generation model.
• Gpt Image 2 2026 04 21 — Image generation model.
• Gpt Image 2.5 Flare — Image generation model.
• Gpt Image 2.5 Flare 2026 09 08 — Image generation model.
• Gpt Image 2.5 Sunburst — Image generation model.
• Gpt Image 2.5 Sunburst 2026 09 08 — Image generation model.

Video models

• Sora 2 — Next-generation video synthesis foundation model from OpenAI producing high-definition video clips with synchronized audio from text and images.
• Sora 2 Pro — Professional-tier OpenAI video generation model delivering high-resolution video clips with synchronized audio and fine-grained camera control.

Audio models

• Gpt 4o Mini Transcribe — Speech-to-text transcription model powered by GPT-4o Mini for fast and accurate audio transcription with language hints.
• Gpt 4o Mini Transcribe 2025 03 20 — Audio model.
• Gpt 4o Mini Transcribe 2025 12 15 — Audio model.
• Gpt 4o Mini Tts — Audio model.
• Gpt 4o Mini Tts 2025 03 20 — Audio model.
• Gpt 4o Mini Tts 2025 12 15 — Audio model.
• Gpt 4o Transcribe — Audio model.
• Gpt 4o Transcribe Diarize — Audio model.
• Gpt Audio — Audio model.
• Gpt Audio 1.5 — Audio model.
• Gpt Audio 2025 08 28 — Audio model.
• Gpt Audio Mini — Audio model.
• Gpt Audio Mini 2025 12 15 — Audio model.
• Gpt Live Transcribe — Audio model.
• Gpt Transcribe — Audio model.
• Tts 1 — Low-latency text-to-speech model from OpenAI optimized for real-time voice generation and interactive conversational applications.
• Tts 1 1106 — Audio model.
• Tts 1 Hd — High-definition text-to-speech model from OpenAI prioritizing audio clarity, natural prosody, and fidelity for broadcast and narration use cases.
• Tts 1 Hd 1106 — Audio model.
• Whisper 1 — Robust automatic speech recognition model from OpenAI supporting multilingual speech transcription, translation, and language identification.

Embedding models

• Text Embedding 3 Large — OpenAI text embedding model producing 3,072-dimensional vector representations across an 8,192-token context window. It scores 64.6 on MTEB and 54.9 on MIRACL for search, clustering, and retrieval systems.
• Text Embedding 3 Small — Cost-effective text embedding model by OpenAI yielding 1536-dimensional vectors with elastic dimension truncation for efficient vector database retrieval.
• Text Embedding Ada 002 — Legacy OpenAI embedding model generating 1536-dimensional embeddings across an 8192-token context window for similarity matching and classification.

Provider clouds

Models in the catalogue are served through these provider clouds. The gateway holds the provider credentials; callers use one nRouter key.

• AWS Bedrock
• Alibaba US
• Anthropic
• Azure Foundry
• Google Vertex AI
• OpenAI
• TypeSafe AI

Product pages

• LLM Gateway — The category page: one OpenAI-compatible endpoint in front of every provider cloud.
• AI Governance — Governance on every request: guardrails, RBAC, audit logs, budgets and SSO.
• OpenAI-Compatible API — The OpenAI SDK contract kept intact — one API for every provider, with routing, guardrails and budgets.
• LLM Cost Optimization — Cut spend with cost-aware routing, caching, exact list pricing and hard budgets.
• Enterprise LLM Gateway — The managed gateway product for production and enterprise workloads.
• Routing — Direct model selection or nrouter/auto, with configured fallback chains.
• Auto — Automatic routing: send nrouter/auto and the platform chooses a model for each request.
• Cache — Exact-match response caching, on by default for non-streaming requests.
• Guardrails — Inline screening for PII, prompt injection and blocked content before egress.
• Virtual Keys — Scoped API keys with per-key budgets, rate limits and model access.
• Budgets — Spend ceilings per key, team and organization, enforced before the provider call.
• Analytics — Cost, usage and performance reporting with CSV export.
• Prompts — Versioned prompt templates with A/B testing across variants and role-based access.
• Observability — Request logs and per-step traces for every call through the gateway.
• Teams — Organizations, teams and role-based access for shared accounts.
• MCP Gateway — Managed access to MCP servers through the same key and controls.
• Playground — Browser-based interactive API testing with your own virtual key.

Use cases

• RAG pipelines — Embeddings and chat through one key, with per-stage cost attribution.
• AI agents — Tool-calling workloads with budgets, guardrails and failover per step.
• Customer support — High-volume support automation with spend ceilings and audit logs.
• Code generation — Coding models routed by task, with cost tracked per repository or team.
• Document processing — Extraction and classification pipelines across document types.
• Voice — Speech and transcription models behind the same endpoint.

Documentation

• Documentation home — API reference, SDKs in every supported language, framework integrations and step-by-step guides, including routing, caching, headers, limits, error handling and getting support.