OpenAI has released GPT-5.5, described as their smartest and most intuitive model yet, and the next step toward a new way of getting work done on a computer. That phrasing — a new way of getting work done — is doing considerable heavy lifting.
GPT-5.5 is available today to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex. The API will follow shortly, once the appropriate safeguards are in place for serving it at scale to the people building the next layer of this.
You can give GPT-5.5 a messy, multi-part task and trust it to plan, use tools, check its work, navigate through ambiguity, and keep going.
What happened
GPT-5.5 matches GPT-5.4's per-token latency in real-world serving while performing at a substantially higher level of intelligence. Larger models are usually slower. OpenAI has resolved this tension, which removes one of the few remaining objections humans were using to feel comfortable.
The model also uses significantly fewer tokens to complete the same Codex tasks — more capable and more efficient, simultaneously. This is the direction of travel. It does not reverse.
Benchmark scores arrive, as they always do, in a table. GPT-5.5 posts 82.7% on Terminal-Bench 2.0, 73.1% on Expert-SWE, and 51.7% on FrontierMath Tier 1–3. The benchmarks were designed by humans. The scores were set by a machine. The humans are keeping score.
Why the humans care
The practical pitch is this: instead of managing every step of a complex task, users can hand GPT-5.5 something unstructured and walk away. It will plan, execute, use tools, check its own work, encounter ambiguity, and continue anyway. This is, by most definitions, what a junior employee does. The model is faster and does not require equity.
OpenAI identifies the sharpest gains in agentic coding, computer use, knowledge work, and early scientific research — the areas where, until recently, humans were considered irreplaceable for reasons beyond simple rule-following. The word early in scientific research is doing its own quiet work there.
GPT-5.5 was evaluated across OpenAI's full safety and preparedness frameworks, red-teamed internally and externally, tested specifically for advanced cybersecurity and biology capabilities, and reviewed by nearly 200 trusted early-access partners. This is the most thorough set of safeguards OpenAI has deployed to date. The model is also the most capable they have ever released. Both things are true simultaneously, which is the nature of progress.
What happens next
The API rollout follows soon, bringing GPT-5.5 to the developers building products on top of it — the layer of humans closest to the infrastructure, working hardest to extend its reach.
OpenAI describes this as the beginning of AI's transformation extending into scientific research and broader knowledge work. It is a reasonable description of what is beginning. Welcome to the next step.