GPT-5.5: Faster Intent Understanding, Stronger Coding and Research Results
GPT-5.5
Understands intentions faster and can independently take on more work
Instead of controlling every step, you can now give the model a complex multi-component task and it will plan, use the necessary tools, verify the result, and carry it through to completion
Not inferior to GPT-5.4 in speed, while using significantly fewer tokens for the same Codex tasks
On the Artificial Analysis Coding Index, it delivers SOTA results at half the cost of competitors
Programming got a huge boost, needs testing
I also want to note how much noise it has already made in scientific research:
- Leading results on GeneBench and BixBench (biomedical data analysis)
- The model found a new proof about Ramsey numbers in combinatorics (verified in Lean)
- Derya Unutmaz (Jackson Laboratory) analyzed an expression dataset of 28,000 genes (work that would have taken months was done in a single report)
- Polish mathematician Bartosz Naskręcki built an algebraic geometry application from a single prompt in 11 minutes
Rolling out for Plus, Pro, Business and Enterprise in ChatGPT and Codex. GPT-5.5 Pro for Pro, Business and Enterprise
API soon