Assess GPT-5.2 coding proficiency using SWE-Bench Pro
SWE-Bench Pro evaluates GPT-5.2's real-world software engineering capabilities across multiple languages.
A model is given a code repository and must generate a patch to solve a realistic software engineering task. SWE-Bench Pro tests four languages and aims to be more contamination-resistant, challenging, diverse, and industrially relevant than SWE-bench Verified.
GPT-5.2 Thinking sets a new state of the art of 55.6% on SWE-Bench Pro. On SWE-bench Verified, GPT-5.2 Thinking scores a new high of 80%.
This translates to more reliable debugging, feature implementation, and code refactoring with less manual intervention.
SWE-Bench Pro is a benchmark for evaluating real-world software engineering tasks. GPT-5.2 Thinking achieves state-of-the-art performance, indicating improved reliability in debugging, implementing features, and refactoring code with reduced manual effort.
SWE-Bench Pro challenges models with realistic software engineering tasks across multiple languages, providing a robust measure of their ability to generate effective code patches.
- Introducing GPT-5.2 ↗OpenAI · OpenAI News · Dec 10, 2025 · paraphrased
First seen on Subvolts Dec 10, 2025. Updated Sep 11, 2026. Spotted a problem? Tell us.
- Introducing GPT-5.2OpenAI · OpenAI News · Dec 10, 2025
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- Model pickGPT-5.2 Thinking for professional knowledge workGPT-5.2 Thinking is the best model for real-world professional use, outperforming experts on knowledge tasks. · 5 min
- HeuristicModel reasoning improves with more thinking timeLLMs perform better on complex tasks when given more time to process and generate a chain of thought. · 15 min
- Model pickUse o1 for advanced reasoning tasksOpenAI o1 model excels in competitive programming, math Olympiads, and complex science benchmarks. · 5 min
- Code snippetGPT-5 custom tool calling with plaintext and grammarsEnable GPT-5 to call tools using plaintext and constrain with developer-supplied context-free grammars. · 5 min
- ToolLife Sciences Research and NGS Analysis pluginsPlugins extending GPT-Rosalind for practical, repeatable scientific workflows in life sciences. · 5 min
- Model pickGPT-5 sizes for performance, cost, and latency trade-offsChoose from gpt-5, gpt-5-mini, and gpt-5-nano for flexible performance, cost, and latency. · 5 min
Questions people ask
- Which AI tools does this work with?
- The sources used it with ChatGPT.
- Where did this come from?
- OpenAI (OpenAI News). Subvolts collected, normalized and credited it; the link above opens the original.