Version of GPT-5.1 Codex optimized for long-running tasks.
GPT-5.1-Codex-Max is OpenAI’s frontier agentic coding model, purpose-built for long-running software engineering tasks that span hours or even days. Announced on November 18–19, 2025, it is the first OpenAI model natively trained to operate across multiple context windows through a process called compaction — coherently working over millions of tokens in a single task by automatically pruning history while preserving the most important context. It replaces GPT-5.1-Codex as the default model in Codex surfaces (CLI, IDE extension, cloud sandbox, and code review).
Built on an update to OpenAI’s foundational reasoning model, GPT-5.1-Codex-Max was trained on real-world software engineering tasks: PR creation, code review, frontend coding, Q&A, and — for the first time — Windows environment operation. The model introduces xhigh (‘Extra High’) reasoning effort, which allocates even more thinking time for non-latency-sensitive tasks. Compaction enables autonomous sessions of 24+ hours, automatically compacting as the context window fills and repeating until the task completes.
From a TheRouter operator’s perspective, GPT-5.1-Codex-Max is listed as active and available through the standard /v1/chat/completions endpoint with full OpenAI-compatible parameters (temperature, max_tokens, tools, tool_choice, response_format, reasoning, stop). It shares GPT-5.1’s 400K context window and 128K max output, with pricing at $1.25/M input tokens, $10/M output tokens, and $0.125/M cached reads. The model is best used for agentic coding workflows routed through Codex or Codex-like harnesses — for general reasoning tasks, route to openai/gpt-5.1 instead.
| Type | Rate |
|---|---|
| Input | $1.35 / 1M tokens |
| Output | $10.80 / 1M tokens |
| Release date | 2025-11-18openai.com ↗ | verified |
| OpenAI API model id | gpt-5.1-codex-maxopenai.com ↗ | verified |
| Model type | Frontier agentic coding model with native compaction — first OpenAI model trained for multi-window context operationopenai.com ↗ | verified |
| Reasoning type | Adaptive — supports ‘none’, ‘low’, ‘medium’, ‘high’, and ‘xhigh’ (Extra High) reasoning effort; xhigh allocates maximum thinking time for non-latency-sensitive tasksopenai.com ↗ | verified |
| Compaction | Native multi-window compaction — automatically prunes session history while preserving critical context, enabling coherent work over millions of tokens across 24+ hour sessionsopenai.com ↗ | verified |
| Windows environment support | First Codex model trained to operate in Windows environmentsopenai.com ↗ | verified |
| Cybersecurity capability (Preparedness Framework) | Most capable cybersecurity model deployed by OpenAI to date — does not reach High threshold but rapidly evolving; treated as High for biology; dedicated cyber monitoring activeopenai.com ↗ | verified |
| License | Proprietary (OpenAI API terms of service / ChatGPT subscription) | verified |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
SWE-bench Verified xhigh reasoning effort, n=500 — vs GPT-5.1-Codex (high) at 73.7%. 30% fewer thinking tokens than GPT-5.1-Codex at same ‘medium’ effort. | 77.9%% | openai.com ↗ | |
SWE-Lancer IC SWE xhigh reasoning effort — vs GPT-5.1-Codex (high) at 66.3%. | 79.9%% | openai.com ↗ | |
Terminal-Bench 2.0 xhigh reasoning effort — vs GPT-5.1-Codex (high) at 52.8%. | 58.1%% | openai.com ↗ |
Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.
curl https://api.therouter.ai/v1/chat/completions -H "Content-Type: application/json" -H "Authorization: Bearer $THE_ROUTER_API_KEY" -d '{
"model": "openai/gpt-5.1-codex-max",
"messages": [
{"role": "user", "content": "Summarize the key points from this input."}
]
}'GPT-5.1-Codex-Max is accessed via the standard chat completions endpoint. For agentic coding workflows, route through Codex or Codex-compatible harnesses to leverage apply_patch and shell tools.
curl https://api.therouter.ai/v1/chat/completions \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"openai/gpt-5.1-codex-max","messages":[{"role":"user","content":"Write a Python script that monitors CPU usage and logs to a file"}],"reasoning_effort":"medium"}'GPT-5.1-Codex-Max becomes the default Codex model, introducing native context compaction for multi-window sessions of 24+ hours, xhigh reasoning effort delivering 77.9% on SWE-bench Verified, and 30% fewer thinking tokens at medium effort vs the predecessor. First Codex model with Windows environment support.
GPT-5.1-Codex-Max is the upgraded default Codex model with three key differentiators: (1) native context compaction for 24+ hour autonomous coding sessions, (2) xhigh reasoning effort for non-latency-sensitive tasks, and (3) significantly higher benchmark scores (SWE-bench Verified 77.9% vs 73.7%). It also uses 30% fewer thinking tokens at the same medium reasoning effort. The model is available via TheRouter API at the same pricing as GPT-5.1-Codex ($1.25/M in, $10/M out).
OpenAI announced GPT-5.1-Codex-Max for Codex surfaces first (CLI, IDE extension, cloud, code review) with API access coming soon. TheRouter lists openai/gpt-5.1-codex-max as an active, available model accessible through the standard /v1/chat/completions endpoint. This means you can route through TheRouter today for API-based workflows, even while Codex-native features (compaction, apply_patch, shell) require Codex-compatible harnesses.
OpenAI explicitly recommends against using Codex-family models for general reasoning or non-coding chat. GPT-5.1 (openai/gpt-5.1) is the general-purpose variant better suited for these workloads at the same price. Use Codex-Max exclusively for agentic coding workflows through Codex or Codex-like harnesses.
| source | URL | retrieved | |
|---|---|---|---|
| Release date | openai.com ↗ | 2026-05-28 | verified |
| OpenAI API model id | openai.com ↗ | 2026-05-28 | verified |
| Model type | openai.com ↗ | 2026-05-28 | verified |
| Reasoning type | openai.com ↗ | 2026-05-28 | verified |
| Compaction | openai.com ↗ | 2026-05-28 | verified |
| Windows environment support | openai.com ↗ | 2026-05-28 | verified |
| Cybersecurity capability (Preparedness Framework) | openai.com ↗ | 2026-05-28 | verified |
| License | — | — | verified |
| SWE-bench Verified | openai.com ↗ | 2026-05-28 | verified |
| SWE-Lancer IC SWE | openai.com ↗ | 2026-05-28 | verified |
| Terminal-Bench 2.0 | openai.com ↗ | 2026-05-28 | verified |
| OpenAI announces GPT-5.1-Codex-Max with compaction and xhigh reasoning | openai.com ↗ | 2026-05-28 | verified |
| How is GPT-5.1-Codex-Max different from GPT-5.1-Codex? | openai.com ↗ | 2026-05-28 | to verify |
| Is GPT-5.1-Codex-Max available via the OpenAI API, or only in Codex? | openai.com ↗ | 2026-05-28 | to verify |
| Can GPT-5.1-Codex-Max be used for general reasoning tasks? | openai.com ↗ | 2026-05-28 | to verify |