OpenAI’s GPT-5.5 release is a model announcement, but the more important signal is product direction. The company is framing the model around long-running computer work: coding, debugging, research, documents, spreadsheets, tool use, and tasks that require checking work across multiple steps.
OpenAI published GPT-5.5 on April 23, 2026, then updated the release on April 24 to say GPT-5.5 and GPT-5.5 Pro were available in the API. That API update matters because it turns the launch from a ChatGPT/Codex experience into infrastructure other developers can build against.
What changed
OpenAI says GPT-5.5 improves over GPT-5.4 while matching GPT-5.4 per-token latency in real-world serving. It points to gains on coding and agent benchmarks, including Terminal-Bench 2.0 and SWE-Bench Pro, and describes stronger performance across knowledge-work evaluations such as GDPval and OSWorld-Verified.
The pricing also signals a new tier of work. OpenAI said gpt-5.5 would be priced at $5 per million input tokens and $30 per million output tokens, while gpt-5.5-pro would be priced at $30 and $180. Those numbers put the model in the category of systems used for expensive work where reliability and completion quality matter.
Why this matters
The old model-release story was mostly about “better answers.” GPT-5.5 is being sold as “more of the work can be handed off.” That changes how teams should evaluate it.
Benchmarks still matter, but the practical test is whether GPT-5.5 can stay inside a task, use tools safely, recover from ambiguity, and produce artifacts that people can review. That is especially relevant for Codex, where the model is expected to operate inside real repositories instead of only responding in a chat box.
What to watch next
Watch API adoption, cyber-safeguard friction, and whether teams measure fewer retries or lower supervision cost. The strongest sign of product-market fit will not be a benchmark; it will be whether GPT-5.5 can make long-running workflows feel less brittle.
