OpenAI recently officially opened its most powerful programming model, GPT‑5.1‑Codex‑Max, to developers through APIs. It focuses on long-running and “agent-style” coding tasks, and its results in a number of authoritative benchmark tests are significantly better than the regular version of GPT‑5.1‑Codex. What’s more attractive is that despite the obvious performance improvement, the pricing of this new model in the API is consistent with GPT‑5, and is regarded by the outside world as a strategic upgrade of “high performance without price increase”.
The report pointed out that GPT-5.1-Codex-Max is trained through a technology called "compaction" and can continue to work in an ultra-long context window of more than one million tokens. Internal tests show that the model can stably perform tasks that last more than 24 hours. In benchmark tests for real software engineering scenarios such as SWE‑Bench Verified, SWE‑Lancer IC SWE and TerminalBench 2.0, the scores of GPT‑5.1‑Codex‑Max increased to 77.9%, 79.9% and 58.1% respectively, which is a significant improvement compared to the regular version of GPT‑5.1‑Codex, highlighting its advantages in complex, multi-step development tasks.
At the commercial level, OpenAI has integrated GPT‑5.1‑Codex‑Max into the API at the same price as GPT‑5, which is approximately US$1.25 per million tokens on the input side and US$10 per million tokens on the output side, while continuing to use the rate limiting policy of GPT‑5. Previously, this model was only open to high-end subscribers using Codex CLI, IDE extensions, cloud services and code review functions, such as ChatGPT Plus, Pro, Business, Edu and Enterprise. Now that it is popularized through the API, OpenAI also announced that it will be set as the default code model in the Codex system.
Unlike the previous Codex series, which tended to be optimized for Unix environments, GPT‑5.1‑Codex‑Max was specifically adapted for the Windows environment during the training phase, making it more directly attractive to programmers who use Windows as their primary development platform. Since the new model is superior to the regular version of GPT-5.1-Codex in terms of speed and tokens utilization efficiency, the report believes that if Windows developers have no special needs, switching to this model is almost a "no-brainer choice."
With this API liberalization, GPT‑5.1‑Codex‑Max has quickly entered the ecosystem of many popular development tools, including coding assistant products such as Cursor, GitHub Copilot, and Linear. In the GitHub Copilot system, the model will be open to Pro, Pro+, Business, and Enterprise subscribers, and will appear in the model selector in multiple usage scenarios, such as the chat, question, edit, and agent modes of Visual Studio Code, as well as portals such as Copilot Chat and Copilot CLI on the GitHub web and mobile terminals.
On a competitive level, OpenAI's move is clearly a response to the strong expansion of rival Anthropic, which has focused its efforts on code models for enterprises and developers in recent years. Its Claude Code tool has become one of the fastest SaaS products to reach the billion-dollar milestone. The report quoted industry expectations as saying that Anthropic is expected to achieve revenue of US$8 to 10 billion this year solely from its code model business for development and enterprises, which also prompted OpenAI to launch GPT‑5‑Codex in August, and in a short period of time, the usage of Codex increased by more than 10 times, and the scale of tokens served every week reached trillions.
In this context, GPT‑5.1‑Codex‑Max is regarded as another important “addition” for OpenAI in the developer market, aiming to strengthen its positive competitiveness with Anthropic in the field of enterprise-level coding tools. However, it is still difficult to judge whether the launch of this new model and its deep integration with GitHub Copilot will be enough to significantly slow down Anthropic's rapid growth in the enterprise coding market. The competition between the two companies in the high-end developer tool track will continue to heat up in the future.
learn more:
https://github.blog/changelog/2025-12-04-openais-gpt-5-1-codex-max-is-now-in-public-preview-for-github-copilot/