Use any model with one API key.
1,000+ models, 0% fee. Up to 70% off tokens. Open source.
import osfrom openai import OpenAIclient = OpenAI( base_url="https://gateway.respan.ai/v1/", api_key=os.environ["RESPAN_API_KEY"],)response = client.chat.completions.create( model=, messages=[ {"role": "user", "content": "Hi"}, ],)print(response.choices[0].message.content)
claude-opus-5-5claude-sonnet-5-5
qwen3.8-max-prime
glm-5.3-prime
gpt-6-lunagpt-6.1-solgpt-6.1-terragpt-6-solgpt-6-astragpt-5.6-solgpt-5.6-terragpt-5.6-lunagpt-5.5claude-haiku-5-5claude-fable-5-1claude-opus-5claude-sonnet-5claude-haiku-4-5
gemini-4-argongemini-3.8-flashgemini-3.5-flash-litegemini-3.1-pro-preview
grok-4.7grok-4.6grok-4.3
deepseek-v4.1-flashdeepseek-v4.1-flash-1017deepseek-v4-proqwen3.8-maxqwen3.7-plusqwen3.8-flashglm-5.3glm-5.3-flash
kimi-k3kimi-k2.6
minimax-m3.1-flashminimax-m3No match. Browse all 1,000+ models
Browse all 1,000+ modelsclaude-opus-5-5
Bulk rates, passed on. 0% fee.
When a provider fails, the next one answers.
claude-opus-5-5429 · 3 tries · 0.85sgpt-6.1-sol200 · 0.96sglm-5.3-primestandbySplit traffic across models, providers or keys.
Anthropic20%
Amazon Bedrock10%Vertex AI10%Primary key20%Overflow key10%Batch key10%Google20%Masks personal data before the model sees it.
Refund order 8841 for Jane Doe, card 4242 4242 4242 4242, receipt to jane@acme.com.
Refund order 8841 for , card , receipt to .
3 masked · name, card, emailnone reached the model
Memory, web search, code and MCP.
qwen3.8-max-prime4 tool callsCalls stop at each customer's budget.
Yes. Claude Code speaks the Anthropic Messages API, and the gateway serves that API for every model, translating streaming, tool calls, and reasoning. Set ANTHROPIC_BASE_URL to the gateway and pick a GPT or Gemini model. To keep Claude and log every call instead, run respan-gateway proxy --anthropic in front of Claude Code.
Respan's gateway is one. OpenRouter is a hosted service you cannot run yourself; Respan's gateway is open source under MIT and Apache-2.0 and runs as a single binary on your own machine or cluster, calling OpenAI, Anthropic, Gemini, xAI, DeepSeek, Mistral, Groq, Ollama, vLLM, and more with your own keys. If you want it hosted after all, the same gateway runs managed on Respan with 1,000+ models on one bill.
For the proxy, yes. It does the same job, one OpenAI-compatible endpoint for every vendor, as a single Rust binary instead of a Python server, and it also serves the Anthropic Messages and Gemini APIs natively, translating between all of them. Routing, fallback, retries, hedged requests, and per-call cost tracking are built in.
In our benchmarks it adds 0.07 ms of latency at the median, serves 56,734 requests per second, and uses 31 MB of memory. On the same setup Bifrost added 0.16 ms at 18,182 requests per second and 488 MB, Portkey 0.17 ms at 4,187 per second and 350 MB, and LiteLLM with one worker 4.34 ms at 268 per second and 418 MB.
Yes. List MCP servers in gateway.toml, over HTTP or stdio, and the gateway serves each one at /mcp/{name} behind the same client keys as everything else, with bearer, OAuth, or one-time login auth. It can also run those tools inside chat requests, so any model can use them from a plain chat call.
Watch film“We scaled from 5M to 500M+ monthly API calls quickly. Respan gave us the debugging layer to resolve production issues 10x faster.”Read story
“Imagine jumping to a log immediately after every LLM call. This is the dream for debugging.”
Watch film“Respan has been key in helping us scale to hundreds of millions of requests with reliable observability into our LLM calls and failure rates.”Read story