OpenAI Unveils GPT-5 and Revolutionary Responses API That Maintains Reasoning Across Conversations
Summary
OpenAI releases GPT-5 alongside groundbreaking Responses API that maintains reasoning across conversations, featuring hosted tools like web search and image generation with 40-80% better cache performance and multimodal capabilities that preserve the model's thinking process between interactions.
Key Points
- OpenAI launches the Responses API alongside GPT-5, designed as an agentic loop that preserves reasoning state across turns unlike Chat Completions which drops reasoning between calls
- The new API offers hosted tools including web search, image generation, and code interpreter, with 40-80% better cache utilization and 5% improvement on TAUBench compared to Chat Completions
- Responses API provides multimodal support from the ground up and emits structured output items showing both what the model said and did, while keeping the model's raw chain-of-thought hidden for safety reasons