Researchers tested more than 20 models and found their estimates of question difficulty aligned poorly with real student data, a gap they call the curse of knowledge.
A Wired column says Alibaba's open-weight Qwen models gained ground in real products while GPT-5 and Llama 4 disappointed, pointing to a shift in the AI balance for 2026.
Moonshot AI released Kimi K2 Thinking as an open-weight model on November 6, claiming state-of-the-art scores on Humanity's Last Exam and BrowseComp.
A public experiment by Nof1 gave GPT-5, Claude, Gemini, DeepSeek, Qwen and Grok $10,000 each to trade crypto perpetuals. Early results put DeepSeek ahead and GPT-5 and Gemini at the bottom.
Independent evaluators at METR, Apollo Research and Haize Labs describe what frontier models can now do, and why they still worry, in a New York Times report by Stephen Witt.
OpenAI showcased a cancer patient consulting GPT-5 at its launch, days after doctors reported a man poisoned by following ChatGPT's diet suggestion. The gap between the two is the story.
