OpenAI has launched GPT-5 and its variants (Pro, Mini, Nano), calling it its “best AI yet” with fewer errors, better coding, and ‘safe completions’ for sensitive queries.
Key points:
- Free access for all ChatGPT users (with limits); Plus and Pro get higher or unlimited use. GPT-5 Pro replaces o3-pro.
- Simulated reasoning now available to free users for improved logical/analytical answers.
- Capabilities: best coding model (74.9% SWE-bench), strong in math, multimodal, medical queries (with disclaimers).
- Accuracy: up to 80% fewer factual errors vs. o3; reduced sycophancy from 14.5% to <6%.
- New features: custom chat colors, preset personalities, Gmail/Google Calendar/Contacts integration (Pro), unified “Advanced Voice,” and safe completions instead of hard refusals.
- Developer API: 256k-token context, cheaper Mini/Nano tiers, new tools like free-form function calling and reasoning effort control.
- Rollout: replaces GPT-4o, o3, o4-mini, 4.1, 4.5; enterprise/education access next week; old voice mode retired in 30 days.
Opinion :
Fascinating.
Note to children - You need to run your life like a GPT π
No comments:
Post a Comment