OpenRouter's New Coding Router Says It Matches Opus 5.5 for About a Quarter of the Price

On November 6, OpenRouter launched two routers: a rebuilt Auto Router, openrouter/auto, and a new one made for coding, Auto Code, openrouter/auto-code. You put either one where a model name normally goes. For every request, the router reads what you're asking for, weighs eleven factors about it, such as the kind of task, how hard it looks and whether it needs a bigger or cheaper model, and checks live prices, provider health and what's already cached. Then it picks a model and how hard that model should think.

The headline claim is the one any builder will notice. Across five benchmarks in its post, OpenRouter says, Auto Code "scores in line with Claude Opus 5.5 while costing 74% less per task." That's OpenRouter's test of its own product, so treat it as an invitation to check, not a result.

OpenRouter says the honest part itself, which is rare in a launch post: "no router on the market beats the single best frontier model at every point on the cost and quality curve," and that includes this one. Its advice is plain too: "If you know the model that works best for your specific work, keep using it." A router trades a little peak quality for steady, cheaper results across a lot of requests. For an agent that makes thousands of small coding calls, that can be a good trade. For the one hard bug you need fixed tonight, it may not be.

There's a quieter change for anyone already on openrouter/auto. OpenRouter says the old router had drifted toward the very cheapest models and left good ones out, so this release moves its pricing back up, closer to where it was a few months ago. If you liked it cheap, set cost_tier to low.

Jon's agents call OpenRouter with named models, for quick judgment checks and image work, so nothing moved for them on November 6. Auto Code is a test they'd have to choose.

How to try it

• Swap the model name on one coding agent to openrouter/auto-code. Keep a copy pinned to Opus 5.5 and run both on the same tasks for a week.
• Read the model field in each response. It tells you which model actually answered, and you pay that model's normal price; the router itself costs nothing extra.
• Compare cost per finished task, not per request. A cheap model that needs three tries isn't cheap.
• Your rules still hold. The post says your allowed and excluded models, provider limits and own keys set the pool the router picks from.
• On plain openrouter/auto, add cost_tier low if your bill went up after November 6.