OpenAI has released ChatGPT 5.5
The focus is on multi-step tasks: planning, tool usage, and self-checking. Key improvements in agentic coding, computer use, knowledge work, and early scientific research. The entire focus shifts to solutions for Codex, Claude Code, etc.
According to OpenAI's benchmarks:
- Terminal-Bench 2.0 increased from 75.1% in GPT-5.4 to 82.7%,
- On the internal Expert-SWE for coding - from 68.5% to 73.1%.
ChatGPT is available in two versions: GPT-5.5 Thinking for Plus, Pro, Business, and Enterprise, GPT-5.5 Pro - only for Pro, Business, and Enterprise. Both models will appear soon in the API, with a price higher than 5.4 ($5/$30 per million tokens), but the model, according to OpenAI, is much more efficient in terms of tokens.
Interestingly, cybersecurity capabilities have been classified for the first time as "High" according to the preparedness framework - they haven't reached "Critical," but the bar is being raised.