Intelligent coding model optimized for long-horizon, agentic coding tasks.
GPT-5.2-Codex is OpenAI's most advanced agentic coding model of its generation, released December 18, 2025. It is a version of GPT-5.2 further optimized for agentic coding in Codex, including improvements on long-horizon work through native context compaction, stronger performance on large code changes like refactors and migrations, improved Windows environment performance, and significantly stronger cybersecurity capabilities. The model was the third major capability jump in the Codex family, following GPT-5-Codex and GPT-5.1-Codex-Max.
From a TheRouter operator's perspective, GPT-5.2-Codex is notable for three things. First, its native context compaction via the /responses/compact API enables sustained multi-hour coding sessions without hitting context limits — a real production differentiator for large-repo agent workflows. Second, it matches GPT-5.2-Thinking on reasoning benchmarks while adding Codex-specific strengths in terminal-based agentic tasks (64% Terminal-Bench 2.0). Third, it has since been superseded by GPT-5.3-Codex (February 2026), which combines the Codex and GPT-5 reasoning stacks — GPT-5.2-Codex remains a cost-effective choice on TheRouter at $1.25/M input tokens with full tool-calling and vision support.
| Type | Rate |
|---|---|
| Input | $1.89 / 1M tokens |
| Output | $15.12 / 1M tokens |
| Release date | December 18, 2025openai.com ↗ | verified |
| Knowledge cutoff | August 31, 2025simonwillison.net ↗ | verified |
| Architecture | Not publicly disclosed (proprietary OpenAI transformer) | unknown |
| Context compaction | Native support via /responses/compact API — loss-aware compression preserving task-relevant informationplatform.openai.com ↗ | verified |
| Windows support | Native — first Codex model trained for Windows environments with PowerShell understandingopenai.com ↗ | verified |
| Successor | GPT-5.3-Codex (February 5, 2026) — combines Codex and GPT-5.2 stacks, 25% fasteropenai.com ↗ | verified |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
SWE-Bench Pro State-of-the-art agentic coding benchmark spanning 4 languages — GPT-5.2-Codex held SOTA at release (Dec 2025). Resolved by GPT-5.3-Codex (Feb 2026). | 56.4%% | openai.com ↗ | |
Terminal-Bench 2.0 Tests AI agents in real terminal environments — compiling code, training models, setting up servers. | 64.0%% | openai.com ↗ | |
SWE-Bench Verified (Python) Python-only SWE-Bench subset. GPT-5-Codex family targets 4-language SWE-Bench Pro as primary benchmark. | ~80%% | nxcode.io ↗ | |
CVE-Bench Cybersecurity vulnerability discovery benchmark — strongest cyber model from OpenAI at time of release. | 87%% | nxcode.io ↗ | |
AIME 2025 American Invitational Mathematics Examination 2025 — GPT-5.2-Codex matches GPT-5.2-Thinking on reasoning. | 100%% | digitalapplied.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-5.2-codex",
"messages": [
{"role": "user", "content": "Summarize the key points from this input."}
]
}'Standard chat completions — works with the OpenAI SDK by swapping baseURL to api.therouter.ai/v1. Supports all GPT-5.2-Codex parameters including reasoning_effort, tools, and response_format.
curl https://api.therouter.ai/v1/chat/completions \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "openai/gpt-5.2-codex",
"messages": [
{"role": "system", "content": "You are an expert software engineer."},
{"role": "user", "content": "Write a Python function to merge two sorted lists in place."}
]
}'GPT-5.3-Codex combines the Codex and GPT-5.2 reasoning stacks into one model, is 25% faster, and sets new SOTA on SWE-Bench Pro and Terminal-Bench. It was the first model OpenAI's own Codex team used to debug and deploy itself.
OpenAI introduced GPT-5.2-Codex as the most advanced agentic coding model, optimized from GPT-5.2 with native context compaction, Windows support, enhanced cybersecurity capabilities, and SOTA on SWE-Bench Pro (56.4%) and Terminal-Bench 2.0 (64.0%).
GPT-5.2-Codex is a version of GPT-5.2 further optimized for agentic coding within the Codex platform. It adds native context compaction (enabling multi-million-token sessions), native Windows support, enhanced cybersecurity capabilities, and Codex-specific optimizations for long-horizon agentic tasks. On reasoning benchmarks the two models perform similarly, but GPT-5.2-Codex excels on agentic coding benchmarks like SWE-Bench Pro and Terminal-Bench 2.0.
GPT-5.3-Codex (released February 2026) is the successor that combines the Codex and GPT-5.2 reasoning stacks. It is 25% faster, achieves higher scores on SWE-Bench Pro and Terminal-Bench, and can do research, tool use, and complex execution beyond pure coding. On TheRouter, GPT-5.3-Codex costs $5 in / $40 out (vs $1.25 / $10 for 5.2-Codex). Choose 5.2-Codex for cost-sensitive coding agent workloads and 5.3-Codex when you need the frontier capability uplift.
Yes. GPT-5.2-Codex is fully compatible with the OpenAI SDK — just swap the baseURL to https://api.therouter.ai/v1 and use the model ID openai/gpt-5.2-codex. All standard parameters work: chat completions, streaming, tools/function calling, JSON mode, vision, and reasoning_effort. The Codex-specific /responses/compact API endpoint is available through the OpenAI Responses API, not the chat completions SDK.
GPT-5.2-Codex was OpenAI's strongest cybersecurity model at release, achieving 87% on CVE-Bench (vulnerability discovery) and demonstrating real-world capability by assisting a researcher in discovering three previously unknown React Server Components vulnerabilities. However, under OpenAI's Preparedness Framework it does not reach the 'High' cybersecurity threshold. The system card outlines additional safeguards including agent sandboxing and configurable network access.
Yes, GPT-5.2-Codex remains actively served on TheRouter even though the successor GPT-5.3-Codex is available. Use model ID openai/gpt-5.2-codex. It's a cost-effective choice at $1.25/M input tokens for coding agent workloads, especially when the frontier capabilities of GPT-5.3-Codex aren't needed.
| source | URL | retrieved | |
|---|---|---|---|
| Release date | openai.com ↗ | 2026-05-28 | verified |
| Knowledge cutoff | simonwillison.net ↗ | 2026-05-28 | verified |
| Architecture | — | — | unknown |
| Context compaction | platform.openai.com ↗ | 2026-05-28 | verified |
| Windows support | openai.com ↗ | 2026-05-28 | verified |
| Successor | openai.com ↗ | 2026-05-28 | verified |
| SWE-Bench Pro | openai.com ↗ | 2026-05-28 | verified |
| Terminal-Bench 2.0 | openai.com ↗ | 2026-05-28 | verified |
| SWE-Bench Verified (Python) | nxcode.io ↗ | 2026-05-28 | to verify |
| CVE-Bench | nxcode.io ↗ | 2026-05-28 | to verify |
| AIME 2025 | digitalapplied.com ↗ | 2026-05-28 | to verify |
| OpenAI releases GPT-5.3-Codex, successor to GPT-5.2-Codex | openai.com ↗ | 2026-05-28 | verified |
| OpenAI launches GPT-5.2-Codex | openai.com ↗ | 2026-05-28 | verified |
| What is the difference between GPT-5.2-Codex and GPT-5.2? | openai.com ↗ | 2026-05-28 | to verify |
| Should I use GPT-5.2-Codex or GPT-5.3-Codex? | openai.com ↗ | 2026-05-28 | to verify |
| Can I use GPT-5.2-Codex through the standard OpenAI SDK on TheRouter? | openai.com ↗ | 2026-05-28 | to verify |
| What cybersecurity capabilities does GPT-5.2-Codex have? | openai.com ↗ | 2026-05-28 | to verify |
| Is GPT-5.2-Codex still available on TheRouter? | openai.com ↗ | 2026-05-28 | to verify |