Tool Suggestions
OpenRouter Auto Code: Opus-Level Coding at 74% Less Cost
OpenRouter has rebuilt the router that picks which AI model answers your request, and added a second router built for coding. The announcement, published on November 6, 2026, introduces the next generation of Auto Router alongside Auto Code, a router tuned for coding agents and harnesses. OpenRouter says Auto Code scores in line with Claude Opus 5.5 across five coding benchmarks at 74% lower cost per task — but that is OpenRouter's own benchmark result, not an independent test, and the saving is measured against calling Opus 5.5 directly. The question this answers is simple: if you use AI for coding, can you stop choosing models by hand and still get top-tier results for less?
What the rebuilt Auto Router does
The new Auto Router is a rebuild from scratch on a new technical foundation, after three years of running the old one. OpenRouter says it now weighs eleven factors about each request — task type, difficulty, speed needs and the rest — and combines them with live pricing, provider health and cache status. From that it picks both the model and the reasoning effort, from the 600-plus models on OpenRouter, within the limits your guardrails, zero data retention settings, provider restrictions and bring-your-own-key policies allow. OpenRouter says the routing step is 3x faster at the 90th percentile — 0.65 seconds against 1.96 seconds — and up to 4x better on its own benchmarks. The router handles a serious load: OpenRouter says Auto Router sees more than 2.5 trillion tokens a month from over 100,000 people. Existing users of openrouter/auto get the new router with no change: the slug is unchanged.
What Auto Code adds for coding
Auto Code (openrouter/auto-code) is the new router built for coding agents and harnesses, created, OpenRouter says, after feedback that a general-purpose router was the wrong tool inside a coding harness. It weighs measured pass rate, speed and cost per solved task across models and reasoning efforts, and it starts in a higher cost tier than the general router. OpenRouter claims it scores in line with Claude Opus 5.5 across five benchmarks — Deep SWE, SWE-Atlas QA, Terminal Bench 4.0, τ³-bench Banking and MMMU Pro Vision — at 74% lower cost per task. On Deep SWE specifically, OpenRouter reports 74.3% at $1.98 per task. Read those numbers as OpenRouter's own: the benchmarks are published on OpenRouter's model routing benchmarks page alongside other routers and single-model baselines, but no independent lab has replicated them. To try it, set model to openrouter/auto-code in any request, or open Auto Router in the OpenRouter chatroom.
What it costs
There is no routing fee. You pay the standard price of whichever model serves the request, so Auto Code has no single token price of its own — the cost depends on the models it picks. A new Auto Router Beta channel, openrouter/auto-beta, carries the next routing changes first, and they graduate to openrouter/auto once they prove out. The cost_tier, allowed_models and excluded_models settings still work.
The catch: the new default costs more than the old Auto
Here is the honest catch, stated by OpenRouter itself: the new default tier is pricier than the old Auto. OpenRouter says it “returned pricing closer to where Auto was a few months ago”. The 74% saving is against calling a frontier model directly, not against the previous router. If you used Auto mainly because it was cheap, set cost_tier=low to get the old cheap bias back. The routers also only choose from the model pool your policies allow.
Who this is for
Auto Code is for anyone running a coding agent or harness who is tired of picking a model for every task — and for anyone paying frontier prices for routine coding work. The honest test is your own workload: run Auto Code on your real tasks for a week and compare the bill and the pass rate against your current setup. OpenRouter's numbers say you will get Opus-level coding for roughly a quarter of the price; only your own tasks can confirm that.