# llm11 > An LLM router that sends every call to the cheapest model that can answer it, checks the answer, and bills in credits at provider cost. llm11 is an LLM router: it sits in front of any OpenAI- or Anthropic-compatible model call, picks the cheapest model from a pool that can answer it well, and returns the answer. It routes using TypeSafe AI's Jev, a "System One" decision model, with a built-in heuristic fallback when Jev is unavailable or unconfigured. Every request is also checked against what it was actually given, and a failed check escalates once to the strongest model in the pool. Every response carries a receipt: what it cost, what the most expensive model in the pool would have cost, and what was kept. ## What decides the routing The routing decision is made by Jev, TypeSafe AI's System One model: a model that answers fixed, typed questions in a single parallel pass with a calibrated confidence, rather than generating text. llm11 asks it which model in the resolved pool should answer, how critical the request is, and how much checking the answer has earned. The decision engine is a pluggable interface, so when Jev is unconfigured or unreachable llm11's own heuristic backend answers the same questions, and every receipt names which backend decided. Benchmark figures published by TypeSafe AI are vendor claims and are reported as such on https://www.llm11.com/jev. ## Integrate - OpenAI wire format: POST https://www.llm11.com/v1/chat/completions, Bearer token, key prefix llm11_live_ or llm11_sandbox_. - Anthropic wire format: POST https://www.llm11.com/v1/messages (Bearer token or x-api-key). - MCP: POST https://www.llm11.com/api/mcp (JSON-RPC 2.0), single tool `verify_request`. - OpenAPI: https://www.llm11.com/api/v1/openapi.json - Pricing as data: https://www.llm11.com/pricing.json Two additions to the OpenAI request shape. `context` (string): supply it and the groundedness check holds every claim in the answer against it; omit it and that check does not run. `model`: a routing pack (llm11-fast, llm11-balanced, llm11-smart, or `llm11-auto` for the project default) or a specific catalogue model id to pin with no routing. `llm11_models` (string array): an explicit candidate pool for this request, overriding `model` and the project default. ## Routing packs - `llm11-fast`: The cheapest end of the catalogue. Best for high-volume, low-stakes traffic. - `llm11-balanced`: A working mix of cheap and capable models. The sane default. - `llm11-smart`: The most capable models available, priced at whatever that costs. Each pack is a price band resolved live against the OpenRouter catalogue, not a fixed list. ## The verification ladder 0. Schema and rules. Always runs, free, deterministic. 1. Groundedness. Runs when `context` was supplied. 2. Cross-model disagreement. Runs when the request is high criticality. 3. Self-consistency. Runs only when the project opted in, the request is high criticality, and the triage call's own confidence sits in the 0.3 to 0.7 uncertain band. ## What it detects, and what it does not Detects: claims unsupported by source text you supplied; a second independent model materially disagreeing; schema, format and reference violations; answers that are unstable under resampling. Does not detect: free-floating factual error where no source was supplied to check against; an answer that is faithful to sources that were themselves wrong; incorrect tool or function calls. llm11 does not claim to eliminate hallucination and no verdict it returns should be read as a guarantee of correctness. ## Pricing There are no feature tiers. A credit is one US cent of upstream spend: the model token, the verification call, and the triage call, all resold at provider list price with zero markup. The only charge is a flat 5% fee taken once, when credits are purchased. - $5 of credit costs $5.25. - $25 of credit costs $26.25. - $100 of credit costs $105.00. - $500 of credit costs $525.00. At a zero credit balance, requests return 402. There is no overdraft and no surprise invoice. ## Not built yet - Streaming. `stream: true` returns 501 rather than a non-streaming response pretending otherwise. - Prompt-cache-loss accounting. Reported as null on the receipt rather than estimated. - Verification of tool and function calls. - Self-hosting. ## Data handling Metadata is always stored (model, verdict, tokens, cost, latency). Prompt and response text are stored only when a project explicitly enables retention and sets a window. ## Pages - https://www.llm11.com/llm-router - https://www.llm11.com/how-it-works - https://www.llm11.com/verification - https://www.llm11.com/pricing - https://www.llm11.com/cheap-llm-router - https://www.llm11.com/llm-router-with-jev - https://www.llm11.com/no-hallucination-llm-router - https://www.llm11.com/alternatives/openrouter - https://www.llm11.com/docs/quickstart - https://www.llm11.com/docs/migrate - https://www.llm11.com/docs/api - https://www.llm11.com/docs/mcp - https://www.llm11.com/jev - https://www.llm11.com/system-one-router - https://www.llm11.com/what-is-a-system-one-model - https://www.llm11.com/blog - https://www.llm11.com/vs/openrouter-auto (llm11 vs OpenRouter Auto) - https://www.llm11.com/vs/litellm (llm11 vs LiteLLM) - https://www.llm11.com/vs/ramp-router (llm11 vs Ramp Router) - https://www.llm11.com/vs/not-diamond (llm11 vs Not Diamond) - https://www.llm11.com/vs/braintrust (llm11 vs Braintrust) - https://www.llm11.com/vs/langfuse (llm11 vs Langfuse) - https://www.llm11.com/vs/portkey (llm11 vs Portkey) - https://www.llm11.com/vs/helicone (llm11 vs Helicone)