GPT-5.2 Codex
GPT-5.2 Codex by OpenAI is a Codex-native coding agent with native compaction, Windows support, vision input, dual-use cyber capabilities, and 400K context.
Model Overview
Capabilities, design details, and architectural traits
GPT-5.2 Codex - Agentic Coding Model for Defensive Cybersecurity
GPT 5.2-Codex is a version of GPT 5.2 further optimized for agentic coding in Codex. Its defining focus is on sustaining complex, multi-step tasks across large repositories over extended sessions - without losing track when plans change or attempts fail. A sharp jump in cybersecurity capability at this release prompted OpenAI to pair it with an invite-only trusted access pilot for vetted defensive security professionals.
What Makes GPT 5.2-Codex Distinct
| Trait | Detail |
|---|---|
| Native compaction | Compresses context in-session to maintain continuity across extended coding tasks without hitting context limits mid-work |
| Large code transformation scope | Explicitly optimized for large-scale refactors, code migrations, and multi-file feature builds, not just single-file completions |
| Vision in coding sessions | Accepts screenshots, technical diagrams, charts, and UI mockups as input; translates design mocks to functional prototypes directly |
| Windows-native agentic coding | Documented improvement in agentic performance within native Windows environments, a capability introduced in the prior generation and substantially improved here |
| Dual-use cybersecurity positioning | Highest cyber capability of any OpenAI model at release, with deployment paired with a staged rollout, added model safeguards, and a vetted Trusted Access pilot for defensive security work |
| Responses API delivery | Available via the Responses API (not Chat Completions), identical to the model running inside the Codex product |
Trusted Access Pilot
GPT 5.2-Codex introduced a documented invite-only trusted access program for security professionals and organizations conducting authorized red-teaming, vulnerability research, or defensive threat emulation. Qualifying participants receive access to more permissive model capabilities than those in the standard API, reflecting OpenAI's stated intent to pair every capability gain with tighter access controls.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | OpenAI | Anthropic | Anthropic |
| Release Date | December 11, 2025 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | - | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 400K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | $1.75 Best Input Pricing | $4 | $2 |
| Output Pricing | $14 | $20 | $10 Best Output Pricing |
| Modalities | |||
| Inputs | textimage | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 28.5 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | - | - | - |
| Agentic Index | - | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from OpenAI
Other models by OpenAI
Top AI Models
Leading alternatives by intelligence score
