How NVIDIA trained Nemotron 3 to hit Olympiad gold
NVIDIA shared its Nemotron 3 post-training recipe and open weights after hitting gold-medal performance in the 2026 IOI and IMO reasoning contests.
Meta's Muse Wonder Models Rival OpenAI and Google
Meta launched its closed Muse Wonder models, ranking fifth on the Vals Index while cutting per-task run costs below rivals from Google and Anthropic.
Ministral 3 in Orbit: AI on the Altair-1 Satellite
The Altair-1 satellite carries Mistral's Ministral 3 into orbit on NVIDIA edge chips to process data directly in space and cut transmission delays.
Microsoft Adds Anthropic Models to Copilot
Microsoft integrates Anthropic Claude models into Copilot, ending OpenAI exclusivity after customer complaints about Excel and PowerPoint.
Liquid AI Releases LFM2.5-24B-A2B for Local PCs
Liquid AI released LFM2.5-24B-A2B, an open-weight reasoning model with 24B total parameters and 2B active, built to run local agents on 16 GB laptops.
Anthropic replaces Git with in-house version control
With Claude driving daily merges to 30,000, Anthropic replaced Git with Scion, an online version control system built for its massive monorepo.
Alibaba Unveils QwenGyre to Speed Up Agent Training
Alibaba Qwen introduced QwenGyre, an elastic framework that dynamically shifts GPUs and cuts long-horizon AI agent training time by 44% in company tests.
OpenAI Revenue Hits $50B Run Rate, Trailing Estimates
OpenAI reached a $50B run rate in September, trailing $70B investor expectations due to an accounting gap between net and gross cloud revenue.
OpenAI Math Papers Face Pushback Over Brute Force
OpenAI released 722 AI-generated math papers on GitHub, but researchers say extended compute power drove the proofs, not conceptual breakthrough.
Step 5 Preview gets one-week free trial on Nous Portal
Nous Research launches a 7-day free trial for Step 5 Preview. The MoE model features 600B total parameters, active 27B, and a 1M context window.