OpenAI launches GPT‑5.1‑Codex‑Max to scale long‑horizon coding

OpenAI has introduced GPT‑5.1‑Codex‑Max, a new agentic coding model designed for sustained, complex software work, and made it available in Codex across the company’s ChatGPT plans.

The San Francisco firm, led by chief executive Sam Altman, has pitched the model as faster, more intelligent and more token‑efficient for day‑to‑day engineering tasks.

GPT‑5.1‑Codex‑Max is trained to operate coherently over very long horizons using a process OpenAI calls “compaction”, allowing the system to prune history and preserve salient context as it progresses.

In internal runs, the model worked independently for hours—sometimes more than a day—on refactors, iterative debugging and end‑to‑end agent loops.

Performance and efficiency

On the SWE‑bench Verified benchmark, the new model at “medium” reasoning effort has reportedly achieved higher scores than GPT‑5.1‑Codex while using roughly 30% fewer thinking tokens, according to OpenAI’s evaluations.

For non‑latency‑sensitive work, an “Extra High” reasoning setting has been added for longer, more deliberate computation.

Real‑world workflows

• Broader task coverage: training focused on practical jobs such as PR creation, code review, front‑end builds and Q&A, with improved behaviour in the Codex CLI.
• Windows support: GPT‑5.1‑Codex‑Max is the first OpenAI coding model trained to operate in Windows environments.
• Tooling surfaces: the model is available in the Codex CLI, IDE extension, cloud and code‑review experiences, with API access “coming soon”.

OpenAI said the model has performed better on long‑horizon and cybersecurity‑relevant evaluations, though it has not reached “High” capability on Cybersecurity under the company’s Preparedness Framework.

The firm described additional mitigations, including dedicated monitoring, disruption of misuse attempts, and a default secure sandbox where file writes are constrained and network access is off unless explicitly enabled.

OpenAI advised treating Codex as an extra reviewer rather than a substitute for human code review.

GPT‑5.1‑Codex‑Max is available now in Codex for ChatGPT Plus, Pro, Business, Edu and Enterprise customers, with OpenAI stating it will replace GPT‑5.1‑Codex as the default in Codex interfaces.

API access via the Codex CLI with API keys is slated to follow. OpenAI recommended the Codex family for agentic coding tasks rather than general‑purpose use.

Internally, OpenAI reported that 95% of its engineers used Codex weekly and shipped roughly 70% more pull requests since adopting it, attributing gains to the combination of the model and tooling.