On this page
OpenRouter added openrouter/auto-code to its public model catalog on 2026-11-05 at 21:12:01 Asia/Shanghai (13:12:01 UTC). The public models API records created 1793884321. The official model page names the listing Auto Code, and the page FAQ says Auto Code was released on November 5, 2026.
Later that evening, at 21:38:09 Asia/Shanghai (13:38:09 UTC), the listing record for the existing general-purpose router openrouter/auto refreshed. The models API created stamp is 1793885889. The official model page now describes Auto as powered by Jev Router and says it replaces market-spend rankings. Auto was already the general-purpose router in use; what changed in this window is the listed mechanism—not a first-time invention of Auto. The same FAQ dates Auto Router as released on November 5, 2026, matching datePublished and this listing refresh.
DigIntoAI hasn't tested either route yet. The notes below follow the official model pages and the public model list. Nearby notes on other OpenRouter listings, and on coding agents and harnesses, are linked under Related reading.
Auto Code, built for coding agents and harnesses
The official Auto Code page describes it as OpenRouter’s router built specifically for coding agents and harnesses. It uses the same routing engine as the Auto Router, with every request treated as part of an agent session. Model choices are guided by coding-task pass rate, completion time, and cost per solved task.
Across turns it follows edits, test results, and recurring errors. It keeps the current model while work is going well, and it tries more reasoning effort before switching when trouble persists. The default coding tier starts from stronger models and weighs failed code changes more heavily. Existing model and provider restrictions still apply.
The models API lists a context length of 2,000,000 and max_completion_tokens of 65,536 for Auto Code. Pricing fields are recorded as -1 / -1, which matches the FAQ: Auto Code has no fixed price of its own.
Auto, now powered by Jev Router
Auto (openrouter/auto, page name Auto Router) remains OpenRouter’s general-purpose router, meant for apps, chat, and general agents. The official model page now says it is powered by Jev Router. Instead of ranking models by market spend, it does request-aware selection of both the model and the reasoning effort, weighing task difficulty, measured benchmark results, live prices, and prompt caching.
For agent sessions, it follows progress across turns, keeps the same model while work is going well, and adjusts reasoning effort or switches models when the task gets stuck. Existing model and provider restrictions still apply.
The models API also lists a context length of 2,000,000 and pricing -1 / -1. The public list does not publish a max_completion_tokens figure for Auto.
Pricing and context on both routers
Neither Auto Code nor Auto has a fixed price of its own. Each request is billed at the price of the model that actually serves it. Both official model pages say they can route to models with context windows of up to 2,000,000 tokens, and that the context length available on a given request is that of the model that serves it.
The llms.txt notes for both IDs also say they are not currently available through EU in-region routing.
Three IDs, and a different context ceiling on Jev Router
openrouter/auto-code, openrouter/auto, and typesafe/jev-router are separate IDs. Calling typesafe/jev-router directly is not the same as calling openrouter/auto, even though Auto is now described as powered by Jev Router.
Jev Router is listed on its own official model page as TypeSafe: Jev Router. The models API created stamp 1790363560 converts to 2026-09-26 03:12:40 Asia/Shanghai, and the page FAQ says Jev Router was released on September 25, 2026. That earlier listing is background for the Jev Router ID.
The Jev Router page gives a different context ceiling: it can route to models with context windows of up to 1,000,000 tokens, again depending on the model that serves the request. That 1,000,000 figure belongs to the Jev Router page. It is not the 2,000,000 ceiling written on the Auto and Auto Code pages—two different page figures.
The documentation guide has not caught up
OpenRouter’s documentation guide for Auto Router still explains the product through market spend: the aggregate spend of millions of people using OpenRouter, measured over a trailing 7-day window for each task type, then ranked by share of spend. The same guide still says two slugs run that router—openrouter/auto and openrouter/auto-beta—and it does not mention Jev or Auto Code. When that guide does not match a model page, the official model pages and the public model list are the current description.
DigIntoAI status
DigIntoAI hasn't run Auto Code or Auto against live coding tasks or chat sessions. There is no measured mix of underlying models, no sample invoice, and no independent pass-rate or timing figure. Coding-task pass rate, completion time, and cost per solved task are the selection signals named on the Auto Code page.
In short
On 2026-11-05 (Asia/Shanghai), OpenRouter listed openrouter/auto-code around 21:12 as a router for coding agents and harnesses that shares Auto Router’s engine and picks models by coding-task pass rate, completion time, and cost per solved task. Around 21:38, the listing for the existing openrouter/auto refreshed: the official model page now says Auto is powered by Jev Router and replaces market-spend rankings with request-aware choice of model and reasoning effort. Both bill at the price of the model that serves the request and can reach a 2,000,000-token context, depending on that model. typesafe/jev-router remains a separate ID, with its own page ceiling of 1,000,000 tokens. DigIntoAI hasn't tested either route yet.