ChatGPT-5.2/5.3 Model Specification Sheet

Core Specifications

Full Name: GPT-5.2 & GPT-5.3-Codex
Developer: OpenAI
GPT-5.2 Initial Release: December 11, 2025
GPT-5.3-Codex Release: February 5, 2026
Predecessor: GPT-5.1
Model Type: Multilingual generative transformer with agentic capabilities

Technical Specifications

Context Window: up to 400,000 tokens (both GPT-5.2 and GPT-5.3-Codex)
Maximum Output: up to ~128,000 tokens
Knowledge Cutoff: ~August 2025
Processing Architecture: Transformer-based with multi-effort reasoning modes

Performance Metrics

Coding & Agentic Benchmarks:
- GPT-5.3-Codex significantly improves performance on SWE-Bench Pro and Terminal-Bench 2.0 relative to GPT-5.2-Codex
- GPT-5.3-Codex shows stronger OSWorld and GDPval scores, reflecting better long-task and professional capabilities

Multimodal & Input/Output Capabilities

Input Modalities: Text, Code, Images, Tool Inputs
Output Modalities: Text, Code, Structured Data
Agentic Interaction: Models can be interactively steered while executing long tasks

Language Support

Multilingual — supports many world languages
Enhanced translation and cross-lingual reasoning capabilities

Advanced Features

Reasoning Modes: Instant, Thinking, Pro, and extended reasoning for deep tasks
Agentic Operation: Execute multi-step workflows, research tasks, and coding projects
Interactive Coding: Code generation, debugging, deployment assistance

API & Pricing

API Token Pricing: ~ $1.75 per million input tokens and $14 per million output tokens for Codex models
ChatGPT Plans: Tiered access including free, Plus, Pro, Business, and Enterprise

Variants & Ecosystem

GPT-5.2 Instant, Thinking, Pro
GPT-5.2-Codex, GPT-5.3-Codex (agentic coding)

Limitations & Notes

Agentic models optimized for coding; creative or conversational tasks may often still use GPT-5.2 standard variants
Availability may roll out over time across platforms and IDEs

Recent Highlights

GPT-5.2 rolled out broadly for professional tasks and long-context work in late 2025
GPT-5.3-Codex — more powerful and faster coding agent launched early 2026