OpenAI has released GPT-5.2, a new model series designed for professional knowledge work that sets new state-of-the-art benchmarks across multiple domains. The update includes Instant, Thinking, and Pro variants available in ChatGPT and via API, offering significant improvements in coding, long-context reasoning, vision, and tool use.
- GPT-5.2 Thinking performs at or above human expert levels on GDPval, beating professionals on 70.9% of tasks while operating at >11x speed and <1% cost.
- It achieves a new state-of-the-art score of 55.6% on SWE-Bench Pro and 80% on SWE-bench Verified, with reduced hallucination rates compared to GPT-5.1.
- The model demonstrates leading performance on OpenAI MRCRv2 for long-context reasoning, achieving near 100% accuracy on the 4-needle variant up to 256k tokens.
- Vision capabilities are enhanced, cutting error rates roughly in half on chart reasoning and reaching 98.7% on Tau2-bench Telecom.
These improvements enable more reliable execution of complex, real-world tasks end-to-end, such as deep document analysis, agentic coding, and visual interpretation.