Skip to content
GPT-5.1-Codex-Max: What It Was, What It Cost and What Replaced It
Model reviews

GPT-5.1-Codex-Max: What It Was, What It Cost and What Replaced It

GPT-5.1-Codex-Max was OpenAI's Codex model for coding jobs that run for hours, and OpenAI shut it down in its API in July 2026.

8 min read

0

GPT-5.1-Codex-Max was OpenAI's coding model for long jobs that run for hours. OpenAI released it on November 19, 2025 inside Codex, where it replaced GPT-5.1-Codex as the default model. In OpenAI's API it cost $1.25 per million input tokens and $10 per million output tokens, with a 400,000-token context window.

It no longer works there. OpenAI shut it down in the API on July 23, 2026 and names GPT-5.6 Sol as the replacement, and Codex's model list today is the GPT-6 family, led by GPT-6.2 Astra and GPT-6.2 Sol.

Last checked November 17, 2026: the model's status, specs and prices on OpenAI's model page, the shutdown date and replacement on OpenAI's deprecations page, and the current lineup on the Codex models page.

What is GPT-5.1-Codex-Max?

GPT-5.1-Codex-Max was a version of GPT-5.1-Codex that OpenAI describes as "optimized for long running tasks." It ran inside Codex, and OpenAI's launch post called it the company's first model trained to keep working across several context windows. A context window is the amount of text a model can hold in mind at once.

OpenAI aimed it at coding jobs too big for one sitting: reorganizing the code of a whole project, long debugging sessions and agent loops that run for hours. It also said this was the first model it had trained to work in Windows environments.

OpenAI recommended GPT-5.1-Codex-Max, and the whole Codex family, only for coding tasks the AI carries out itself in Codex or tools like it. For everything else it pointed people to GPT-5.1, its general-purpose model.

FactGPT-5.1-Codex-MaxWhere it comes from
ReleasedNovember 19, 2025, in CodexLaunch-day coverage
Where it ran at launchCodex CLI, IDE extension, cloud and code review; API access "coming soon"OpenAI launch post
Where it ran in the APIThe Responses API onlyOpenAI model page
Context window400,000 tokensOpenAI model page
Longest answer128,000 tokensOpenAI model page
Knowledge cutoffSeptember 30, 2024OpenAI model page
API price per million tokens$1.25 input, $0.125 cached input, $10 outputOpenAI model page
Shutdown noticeApril 22, 2026OpenAI deprecations page
Shut down in the APIJuly 23, 2026OpenAI deprecations page
Named replacementgpt-5.6-solOpenAI deprecations page
GPT-5.1-Codex-Max, Shut down, gpt-5.1-codex-max. A version of GPT-5.1-Codex optimized for long running tasks. Reasoning Higher, Speed Fast, Price $1.25 · $10 Input · Output. 400,000 context window, 128,000 max output tokens, Sep 30, 2024 knowledge cutoff. Text tokens: Input $1.25, Cached input $0.125, Output $10.00
OpenAI's API page for the model, with the "Shut down" label beside its name and the per-token prices, as it read on November 17, 2026. Source: OpenAI.

Is GPT-5.1-Codex-Max still available?

No. OpenAI shut GPT-5.1-Codex-Max down in its API on July 23, 2026, after giving notice on April 22, 2026. Its model page now labels both the model and its only snapshot (the fixed version developers call by name) "Shut down", so any code that still asks for gpt-5.1-codex-max needs a new model name.

It is also gone from Codex's model list. The models the Codex page recommends are all GPT-6 models, and the page doesn't mention GPT-5.1-Codex-Max at all.

Its two siblings were shut down on the same day. GPT-5.1-Codex was also replaced by GPT-5.6 Sol, and the smaller GPT-5.1-Codex-mini was replaced by GPT-5.6 Terra.

Jul 23, 2026 gpt-5.1-chat-latest gpt-5.6-sol; Jul 23, 2026 gpt-5.1-codex gpt-5.6-sol; Jul 23, 2026 gpt-5.1-codex-max gpt-5.6-sol; Jul 23, 2026 gpt-5.1-codex-mini gpt-5.6-terra
Four rows from OpenAI's past deprecations table: the shutdown date, the retired model and its replacement. Source: OpenAI.

What replaced GPT-5.1-Codex-Max?

OpenAI names GPT-5.6 Sol (gpt-5.6-sol) as the replacement in its API. Sol was the top model of the GPT-5.6 family, which OpenAI opened to everyone worldwide in July 2026 after a preview limited to US government-approved organizations.

Pat Simmons has put GPT-5.6 Sol on real builds. In a round of four fixed builds against GPT-6 Astra and Claude Fable 5.1 (a 3D kart racing game, a walking machine simulation, a 3D dive watch product page and a rebuild of an award-winning website), Sol was by far the cheapest at $18.42 for all four, against $93.84 for Astra and $429.62 for Fable 5.1. Those are API-equivalent costs, and that round named no winner; every build is live in our write-up of how GPT-5.6 Sol did on four builds.

Pat also reviewed GPT-5.6 Sol on his YouTube channel in "GPT 5.6 Sol: No-Hype Full Review & Testing" (July 10, 2026), listed in the sources below.

The deprecations page points no further than GPT-5.6 Sol. Codex itself has since moved to the GPT-6 family, covered in the last section.

How much did GPT-5.1-Codex-Max cost?

In OpenAI's API, GPT-5.1-Codex-Max cost $1.25 per million input tokens and $10 per million output tokens, the chunks of text a model reads and writes. Cached input, text the model had already been sent recently, cost $0.125 per million.

Inside Codex it came with a paid ChatGPT plan. OpenAI's launch post listed Plus, Pro, Business, Edu and Enterprise, each with its own usage limits.

OpenAI also said the model got more done per token. On SWE-bench Verified at medium effort, it beat GPT-5.1-Codex while using 30% fewer "thinking tokens", the reasoning a model works through before it answers.

What was compaction in GPT-5.1-Codex-Max?

Compaction is how GPT-5.1-Codex-Max kept working once its memory for a session filled up. OpenAI said that in Codex, when a session neared the context-window limit, the model pruned its history while keeping the most important context, which gave it a fresh window, and repeated that until the task was done.

OpenAI said this let the model work over millions of tokens in one task and work on its own for hours. In its internal tests, OpenAI said, it had seen the model work on tasks for more than 24 hours, fixing failed tests until it delivered a result.

Other coding tools already did something similar: Claude Code automatically summarized earlier turns when it ran out of room, and at launch developers asked how different OpenAI's version was. OpenAI's post says the model was trained to compact natively and doesn't spell out more than that.

How did it score on OpenAI's coding tests?

On OpenAI's own figures, GPT-5.1-Codex-Max at its new "xhigh" effort setting scored 77.9% on SWE-bench Verified, a set of 500 coding tasks, against 73.7% for GPT-5.1-Codex at high effort. AI for Mortals never tested GPT-5.1-Codex-Max, so every score here is one OpenAI reported.

Test (OpenAI's own figures)GPT-5.1-Codex (high)GPT-5.1-Codex-Max (xhigh)
SWE-bench Verified (500 tasks)73.7%77.9%
SWE-Lancer IC SWE66.3%79.9%
Terminal-Bench 2.052.8%58.1%

At the regular "high" setting, OpenAI's SWE-bench Verified score was 76.5%. The rival scores published at the time were 77.2% for Claude Sonnet 4.5 and 76.2% for Gemini 3 Pro on that test, and 54.2% for Gemini 3 Pro and 42.8% for Sonnet 4.5 on Terminal-Bench 2.0.

OpenAI added xhigh ("Extra High") for jobs where waiting longer is fine, since the model thinks for longer before it answers. It still recommended medium as the everyday setting.

Is Codex better than ChatGPT?

Codex and ChatGPT do different jobs. Codex is OpenAI's coding agent, which writes to the files in its workspace and runs in the Codex app, the command line, an editor extension and the cloud, and ChatGPT is OpenAI's general assistant.

When it launched GPT-5.1-Codex-Max, OpenAI advised using Codex models only for coding tasks in Codex, with the general-purpose GPT-5.1 for everything else. The line is blurrier now: the Codex models page says GPT-6.2 Astra and Sol run in both ChatGPT Work and Codex, and it suggests Sol for everyday coding and writing alike.

Our guide to what Codex does and what changed covers the app itself, and our company page sets out what OpenAI makes beyond it.

Which Codex model should you use today?

In Codex, OpenAI recommends GPT-6.2 Astra for demanding work, like tracing a bug across services, and GPT-6.2 Sol for everyday coding, writing and work across several sources. Sol costs fewer credits per token than Astra.

Both came out on November 17, 2026. When AI for Mortals gave GPT-6.2 Astra and Sol, tested on five office jobs, every model got every job right, and the advice was to use GPT-6.2 Sol as your everyday model and save Astra for decks, designed documents and hard projects.

Both have more room than GPT-5.1-Codex-Max had. The Codex page lists a default context of 500,000 tokens for GPT-6.2 Sol and 272,000 for GPT-6.2 Astra, and both can go up to 1 million tokens.

If you called gpt-5.1-codex-max through the API, OpenAI's listed swap is gpt-5.6-sol. The Codex page also lists GPT-6.1 Terra as a fast, efficient choice for writing and computer use, and GPT-6 Luna for focused, repeatable tasks.

Recommended models. GPT-6.2 Astra and GPT-6.2 Sol are now available in ChatGPT Work and Codex on supported plans and clients. For demanding work across code, apps, or research, choose GPT-6.2 Astra. Choose GPT-6.2 Sol for everyday coding, writing, and work across several sources. It costs fewer credits per token than GPT-6.2 Astra at Standard speed.
The "Recommended models" section of OpenAI's Codex models page on November 17, 2026. Source: OpenAI.

Questions people ask about GPT-5.1-Codex-Max

What is the difference between GPT-5.1-Codex and GPT-5.1-Codex-Max?

Codex-Max was the version built for long jobs, trained to compact its session and keep going. It scored higher than GPT-5.1-Codex on all three coding tests in OpenAI's launch post and replaced it as Codex's default, and both were shut down in the API on July 23, 2026.

Was GPT-5.1-Codex-Max free?

No. In Codex it came with the paid Plus, Pro, Business, Edu and Enterprise plans, and in the API it was billed per token.

What replaced GPT-5.1-Codex-mini?

OpenAI replaced GPT-5.1-Codex-mini with GPT-5.6 Terra on the same July 23, 2026 shutdown date.

What could GPT-5.1-Codex-Max read and write?

OpenAI's model page lists text and images as input and text as output, with up to 128,000 tokens in a single answer.

Is GPT-5.5 still available?

GPT-5.5 left Codex for people signed in with ChatGPT on October 14, 2026, and the Codex page points Plus, Pro, Business, Enterprise and Edu users to GPT-6.1 Sol and Free and Go users to GPT-6 Luna. It stays available in OpenAI's API and in Codex sessions signed in with an API key.

Sources

What did you think of this post?

Comments

Sign in to join the conversation.

You might also like.

Articles, builds, skills and tools on the same idea.
More At Work

AI FOR MORTALS.

Your daily AI edge that will make everyone wonder how you do it all.

Every weekday, the practical AI updates that matter, and exactly what to do with them.

The latest in AI, in plain English. Free, every weekday.