OpenAI unveiled GPT‑5 on 7 August 2025, touting it as its smartest and most useful model so far. The company said GPT‑5 unifies a default model with a “deeper reasoning” model and a router that decides which to use based on the query. OpenAI claimed that GPT‑5 delivers state‑of‑the‑art performance across coding, math, writing, health and visual tasks and reduces hallucinations. A major change is the introduction of “safe completions,” a fine‑tuning technique that trains the model to produce helpful yet safe answers and to suggest alternative outputs when it must refuse to answer.
However, the rollout sparked backlash from some users. As part of the update, OpenAI removed eight legacy models and replaced the ChatGPT model picker with an automated routing system. Users complained that GPT‑5’s responses were shorter and less nuanced, and many preferred GPT‑4o for creative or emotional tasks. After widespread criticism on social media and subreddit forums, CEO Sam Altman announced that GPT‑4o would be reinstated as an option for Plus subscribers and promised greater transparency regarding which model is in use.
- Google Integrates Ads into AI-Powered Search Experiences
- Hyperledger Fabric 3.1.1 and Fabric-X: Pioneering the Next Generation of Enterprise Blockchain
- Manus AI: China’s Leap into Autonomous AI Agents
- R‑Zero: a self‑evolving reasoning LLM framework
- MIT Spinout Develops AI That Acknowledges Its Own Uncertainty to Combat Hallucinations
- DeepSeek’s R1-0528 Model Raises Concerns Over Free Speech Limitations