OpenAI has released GPT-5 in its API platform, positioning it as a state-of-the-art model for coding and agentic tasks. The release includes three model sizes—gpt-5, gpt-5-mini, and gpt-5-nano—and new parameters for controlling verbosity and reasoning effort.
- GPT-5 scores 74.9% on SWE-bench Verified and 88% on Aider polyglot, outperforming o3 in internal testing.
- It achieves 96.7% on the τ2-bench telecom tool-calling benchmark and excels at long-context retrieval.
- The API introduces a verbosity parameter (low, medium, high) and supports custom tools with plaintext inputs.
- Early feedback from Cursor, Windsurf, and Vercel highlights improved steerability, frontend aesthetics, and agent stability.
The model is designed to serve as a coding collaborator with enhanced tool intelligence for complex, real-world tasks.