OpenRouter's New Auto Router and Auto Code: Pick a Model Per Request, Pay 74% Less Than Opus 5.5

OpenRouter rebuilt its Auto Router from scratch and shipped a second, coding-specialized router alongside it. The announcement landed on November 6, and it matters for anyone who has been hand-picking models for every agent, harness, or app: the new routers choose the model and the reasoning effort for each request, inside the guardrails you set.

Here is what actually changed, what the numbers mean, and when a router is the wrong choice.

Two routers, one engine

• Auto Router (openrouter/auto) — the general-purpose router for mixed traffic, apps, and general agents. It favors a cost-efficient tier of models.
• Auto Code (openrouter/auto-code) — new, built for coding agents and harnesses. It starts in a higher cost tier, weighs measured pass rate, speed, and cost per solved task, and penalizes failures more harshly so it course-corrects faster on work where success and failure are clear-cut.

Both run on the same routing engine. The slug openrouter/auto is unchanged, so existing callers get the new router with no code change.

How routing works now

The old router leaned on a couple of imprecise signals, mostly task type. The new one rates eleven factors about each request in a single fast decision-model call — what kind of task it is, whether the task just changed, how hard it is, whether speed matters, whether it needs vision or knowledge of a codebase, and whether a bigger model, more thinking, or a cheaper model would serve it better — then combines that with live pricing, provider health, and cache status.

A request moves through five steps:

1. Narrow the pool. Guardrails, zero data retention requirements, model capabilities, provider restrictions, BYOK keys, and your allowed/excluded model lists decide which models are even considered.
2. Rate the request. The decision model scores the eleven factors.
3. Pick model and effort. The routing policy weighs live prices, cache status, and benchmark performance.
4. Provider routing. The request goes through the same provider routing a direct request would, landing on the best provider for that model.
5. Learn from the turn. Signals like whether the model is struggling and what is already cached feed into the next turn's decision.

Every one of the 600+ models on OpenRouter enters the consideration set the day it is added; benchmarks and popularity signals decide whether it earns a place in the cost-optimal set. You do not need to track every release yourself.

The numbers OpenRouter published

• Up to 4x better than the old router on benchmarks, depending on the benchmark, with a routing step 3x faster at p90: 0.65s vs 1.96s in OpenRouter's live test.
• Auto Code scores in line with Claude Opus 5.5 across five benchmarks — Deep SWE, SWE-Atlas QA, Terminal Bench 4.0, tau-bench Banking, and MMMU Pro Vision — while costing 74% less per task.
• In beta against the previous router, invalid tool calls were 4x lower and agentic responses were cut off half as often. The router stayed on the same model for nearly nine of every ten agent turns, which protects prompt-cache reuse.
• Scale context: Auto Router sees over 2.5 trillion tokens from over 100,000 people every month.

These are OpenRouter's own figures, published with its announcement, and the benchmark results live on its public model routing benchmarks page, rerun periodically. Treat the 74% figure as a cost-per-task claim across those five benchmarks, not a guarantee for your workload — your mix of tasks decides your actual bill.

Two details worth knowing before you switch

The default got more expensive on purpose. The old router had drifted toward the cheapest models, leaving out many strong ones. OpenRouter deliberately moved default pricing back toward where Auto was a few months ago, to get a greater variety of models into the mix. If you preferred the old behavior, set cost_tier to low.

It is restrained about switching. Switching models costs extra while the new model rebuilds the input token cache, so the router prefers changing reasoning effort when that is enough, and switches only when the gain is worth the cache it gives up. When a session is going wrong in ways that point at the model — the same error coming back, tests failing after an edit — those signals raise the odds of a switch.

Getting started

It works wherever you would normally put a model string:

curl https://openrouter.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openrouter/auto-code",
    "messages": [{"role": "user", "content": "Add a retry with backoff to fetchUser in api/users.ts"}]
  }'

There is no routing fee: you pay the standard price of whichever model serves the request, and the response's model field tells you which one it was. The auto-router plugin settings cost_tier, allowed_models, and excluded_models still work. openrouter/auto-beta is the release channel where new routing changes land before graduating to openrouter/auto.

You can also see the reasoning behind a routing decision: pick Auto Router or Auto Code in the OpenRouter chatroom and a routing insights panel appears beside each reply, showing the task classification, the selection share across candidate models, and the difficulty, precision, and larger-model-benefit judgments. The same analysis is available via the API with router metadata enabled. The Auto Router and Auto Code model pages show a live routing mix — which models each router picks most often, split by session type, and how the mix shifts as difficulty rises.

Model routing vs provider routing

Do not confuse the two. Model routing picks which model handles a request; provider routing picks who serves that model. Auto Router and Auto Code do the first, then OpenRouter's provider routing does the second using your provider preferences, live uptime and latency, price, and whether a provider already holds your prompt cache.

When a router is the wrong choice

OpenRouter says it plainly in the announcement: benchmarks consistently show that no router beats the single best frontier model at every point on the cost and quality curve, and the new Auto Router is no exception. Routers earn their place when the work is varied, when an organization needs to hold costs steady, and for people who do not have the time or interest to keep up with the constant stream of model releases.

If you know the model that works best for your specific work, keep using it. If you would rather choose once and stay focused on the work itself, a router is the way to do that. Auto Code is the more interesting half of this launch for developers: coding harnesses were the most common complaint about previous Auto Router versions, and a router that weighs pass rate and penalizes failures is a genuinely different tool from a general-purpose one.

Source: OpenRouter, "Introducing the New Auto and Auto Code Routers" (November 6, 2026), and the OpenRouter Auto Router documentation.