OpenRouter launches Auto and Auto Code model routers
OpenRouter has launched a redesigned Auto Router and a new Auto Code router for coding work, using live request signals, pricing and provider data to choose models. The company says the new system outperforms its previous router by up to 4× on benchmarks and makes routing decisions 3× faster at the p90 in its live test.
- OpenRouter launched redesigned Auto Router and coding-focused Auto Code.
- The company reports up to 4× improvement over its previous router on benchmarks.
- Auto Code scored in line with Claude Opus 5.5 at 74% lower cost per task in five benchmarks.
- Users can inspect routing decisions in the chatroom or through API router metadata.
- Both routers are available through the model names openrouter/auto and openrouter/auto-code.
A rebuilt routing system
OpenRouter says it rebuilt Auto Router from the ground up after three years of operating the service. The new Auto Router and Auto Code evaluate eleven factors about each request, then combine those signals with live pricing, provider health and cache status.
The system can work with any set of models and supports governance policies such as guardrails, zero data retention requirements, provider restrictions, allowed model lists and BYOK keys. OpenRouter says Auto Router sees more than 2.5T tokens from over 100k people every month, creating a production environment for its routing decisions.
In the company’s live test, the routing step took 0.65s at the p90, compared with 1.96s for the previous router. OpenRouter also reports that the new system outperformed the old router by up to 4×, depending on the benchmark.
Auto Code targets coding harnesses
Auto Code, available as openrouter/auto-code, is tuned specifically for coding work. OpenRouter says it uses a higher tier of coding models, places greater weight on latency and penalizes failures more heavily, allowing it to react when a session produces repeated errors or failing tests.
Across five benchmarks cited by the company, Auto Code scored in line with Claude Opus 5.5 while costing 74% less per task. OpenRouter says the router also limits model switching when a change would sacrifice useful prompt cache data. During its beta period, it stayed with the same model for nearly nine of every ten agent turns.
For the general purpose router, OpenRouter changed its approach to model variety after observing that low-cost models had gradually reduced the cost per task while excluding other models. Users who want behavior similar to the previous cost profile can set cost_tier to low.
Visibility into model choices
Users can try the routers by setting the model to openrouter/auto or openrouter/auto-code in an API request, or by selecting Auto Router in OpenRouter’s chatroom. A routing insights panel explains why a model was selected, and the same analysis is available through the API when router metadata is enabled.
The model pages show which models each router selects most often, separated by session type and showing how the mix changes as task difficulty increases. OpenRouter also publishes benchmark results alongside other routers and single-model baselines, and says it reruns the tests periodically.
The launch gives users a choice between selecting a model directly and delegating that decision to a router. OpenRouter acknowledges that no router consistently beats the best single frontier model across the entire cost and quality curve. It positions Auto Router and Auto Code for varied workloads, controlled costs and users who prefer not to track every model release.
OpenRouter: Introducing the New Auto and Auto Code Routers
