Model page

GPT-5.2

Pinned GPT-5.2 snapshot for stable behavior.

About GPT-5.2

GPT-5.2 is built for the demands of professional work at scale — processing 400,000-token inputs while maintaining precision across long technical documents, entire codebases, and multi-step agentic pipelines. Its AIME 2025 score of 100% and 75.4% on SWE-Bench Verified signal genuine capability in structured reasoning and real-world software engineering, not just benchmark theater. Users consistently praise the quality of its professional deliverables: spreadsheets, financial models, presentations, and complex front-end code that arrives cleaner and with fewer debugging cycles than its predecessors. The GPT-5.2-Codex variant adds security-aware enterprise engineering for long-horizon development projects. That said, users who valued the warmth and creativity of earlier GPT models have noticed something different here — responses lean measured and formal, and can feel truncated where you might expect more depth or nuance. At $1.75 input / $14 output, it is a deliberate choice for high-stakes professional workloads rather than general-purpose conversation. Three variants — Instant, Thinking, and Pro — let teams dial in the right balance of speed and reasoning depth.

Best for

  • Enterprise software engineering with GPT-5.2-Codex, particularly large-scale refactoring, security-aware development, and long-horizon coding projects
  • Professional document generation: financial models, spreadsheets, project plans, and business automation workflows
  • Long-document analysis and knowledge work — processing full codebases, API documentation, or technical specifications up to 400k tokens
  • Complex multi-step agentic workflows requiring consistent tool calling, structured outputs, and reliable instruction-following across many steps
  • Chart and dashboard interpretation, including UI screenshot analysis and vision-heavy business intelligence tasks

Specs & capabilities

How GPT-5.2 stacks up — intelligence, speed, context, and modalities.

Capability

Intelligence

High

Capability

Speed

Medium

Capability

Context window

400,000 tokens

Capability

Max output

128,000 tokens

Capability

Knowledge cutoff

August 31, 2025

Frequently asked questions

What does GPT-5.2 cost?

Input is $1.75 per million tokens and output is $14.00 per million tokens. Cache hits receive a 45% discount, bringing cached input to roughly $0.96 per million tokens. That is about 40% more expensive than GPT-5.1.

How large is the context window?

GPT-5.2 accepts up to 400,000 tokens of input and can generate up to 128,000 tokens of output — a meaningful expansion over its predecessor that makes it practical for whole-codebase or long-document work.

What is GPT-5.2 best at?

Professional knowledge work: coding, debugging, refactoring, financial modeling, and multi-step agentic tasks. It scored 100% on AIME 2025 and 75.4% on SWE-Bench Verified, reflecting its strength in structured, high-stakes reasoning.

What are its notable limitations?

Users report that responses can feel corporate and less creative than GPT-4o — shorter, more formulaic, and less willing to explore nuance. Precise visual tasks (counting objects, generating diagrams) remain weak, and subtle hallucinations can appear in long chained workflows.

Which variant should I choose — Instant, Thinking, or Pro?

Instant is speed-optimized for latency-sensitive tasks. Thinking provides standard reasoning suitable for most professional use. Pro enables extended reasoning for the most demanding multi-step problems. GPT-5.2-Codex is a separate specialized variant for enterprise software engineering.

How does GPT-5.2 compare to GPT-5.1?

GPT-5.2 brings a larger context window (400k vs. smaller), improved coding reliability, better long-document comprehension, and stronger multi-step agentic performance — at a 40% price premium. If your workload does not require the extended context or deeper reasoning, GPT-5.1 may offer better cost efficiency.

Related models