OpenRouter rebuilds Auto Router and adds Auto Code, a model router tuned for coding agents

On November 6, 2026, OpenRouter launched a next generation of Auto Router, rebuilt on a new technical foundation, together with Auto Code, a new router for coding work. You use them by setting the model to openrouter/auto or openrouter/auto-code, and openrouter/auto-beta is the release channel where upcoming routing changes appear first. For each request, a fast decision model rates eleven factors, such as the kind of task, whether the task just changed, how hard it is, whether speed matters and whether it needs vision or knowledge of a codebase. A routing policy combines those ratings with live prices, provider health and cache status to pick both the model and the reasoning effort, and OpenRouter's normal provider routing then decides who serves it.

OpenRouter says the new router outperforms the old one by up to 4× depending on the benchmark, and that its routing step was 3× faster at the p90 in a live test, 0.65 s against 1.96 s. In its beta, compared with the previous router, invalid tool calls were 4× lower and agentic responses were cut off half as often. To protect the prompt cache, the router avoids switching models and in beta stayed on the same one for nearly nine of every ten agent turns. Auto Code favours a higher tier of coding models, gives more weight to speed and penalizes failures more; OpenRouter says that across five benchmarks it scores in line with Claude Opus 5.5 at 74% less cost per task.

There is no router fee: you pay the standard price of the model that serves the request, and the response's model field names it. Existing openrouter/auto users change nothing, and cost_tier, allowed_models and excluded_models still work. OpenRouter moved Auto Router back to pricing closer to where it was a few months ago, after it had drifted toward the cheapest models, so that more models are included; that can mean a higher cost per task than recent months, and setting cost_tier to low restores behaviour like before.

The model pool respects the account's guardrails, zero data retention, provider restrictions, allowed and excluded model lists and BYOK keys. The chatroom shows why each model was picked, the API returns the same through router metadata, model pages show the live routing mix, and results appear on a model routing benchmarks page.