OpenAI's unreleased next-generation model — leaked benchmarks suggest PhD-level reasoning.
GPT-5 is the anticipated next major model from OpenAI, with internal benchmarks allegedly showing a 92% score on the GPQA Diamond dataset (PhD-level STEM problems) — a 34-point jump over GPT-4o. The model reportedly uses a 'long-chain reasoning' architecture, processes a 2M token context window, and is multimodal across text, image, audio, and video by default. Expected to launch Q2 2026 with API access and a ChatGPT integration, GPT-5 is anticipated to be the most capable publicly available AI model when it ships.
Get stories like these every Monday. No noise, just signal.
12,400+ subscribers. Unsubscribe anytime.