Greg Brockman: GPT-5.5 is very good

Greg Brockman used a short X reply on May 23 to make a clear point about OpenAI’s latest flagship model. Responding to comparisons involving Anthropic’s Claude and Opus, the OpenAI co-founder wrote that “GPT-5.5 is a very good model,” a public endorsement that landed in the middle of an active developer debate over which frontier model is best for coding and agent-style work. (openai.com) OpenAI had already put GPT-5.5 at the center of its product push in late April. In its April 23 launch post, the company called GPT-5.5 its “smartest and most intuitive to use model yet” and framed it as a step toward handling real work on computers. (openai.com) ### Why did Brockman’s one-line post get attention? Greg Brockman is not just a commentator on model launches; he is OpenAI’s co-founder and president, and he has been one of the company’s main public voices around GPT-5.5 since its release. (openai.com) On a press call reported by TechCrunch and CNBC on April 23, Brockman described the model as a step toward more agentic computing and stronger performance in coding, computer use and research tasks. The May 23 comment matters because it was not a formal launch claim or benchmark chart. (openai.com) It was a direct response in the social-media stream where developers, founders and researchers often compare Claude, Opus, Gemini and OpenAI models on practical workloads. That makes the post easy to cite in ongoing model-selection conversations, especially when teams are choosing tools for coding assistants or autonomous workflows. This is an inference based on the context of the exchange and the broader comparison market. (techcrunch.com) ### What had OpenAI already said about GPT-5.5? OpenAI said on April 23 that GPT-5.5 was built for “real work” and for tasks including writing code, researching online, analyzing information, creating documents and spreadsheets, and moving across tools. Its system card says the model is designed to understand tasks earlier, ask for less guidance and use tools more effectively. April 24 brought a commercial step as well. OpenAI updated its launch post and system card to say GPT-5.5 and GPT-5.5 Pro were available in the API, extending the release beyond ChatGPT and Codex. (datacamp.com) OpenAI’s API documentation also lists GPT-5.5 pricing, showing the model had moved from announcement to deployable product. ### Which workloads are at the center of these comparisons? Coding and agentic terminal work are the main battlegrounds in recent public comparisons. OpenAI’s launch materials and outside coverage both emphasized GPT-5.5’s performance in coding, computer use and longer multi-step tasks, while comparison write-ups have paired it directly against Anthropic’s Claude Opus 4.7 for production coding and tool use. (openai.com) Benchmark discussions are also part of that debate. OpenAI’s system card and API materials describe GPT-5.5 as a model for complex, real-world tasks, and third-party comparisons have treated it as especially strong in agentic terminal workflows while often giving Claude Opus high marks in multi-file coding. (openai.com) Those outside comparisons are not OpenAI claims, but they help explain why Brockman’s comment resonated. ### What comes next for developers watching this? OpenAI’s next visible milestones are already on the record. (openai.com) GPT-5.5 and GPT-5.5 Pro are in the API, and OpenAI has a GPT-5.5 bio bug bounty open through June 22, 2026, for jailbreak testing tied to Codex Desktop. That means the next round of evidence is likely to come from usage rather than slogans. Developers comparing GPT-5.5 with Claude Opus and other frontier models now have API access, published pricing and public safety documentation to test against their own coding and agent workloads. (openai.com) (developers.openai.com) (openai.com)