Today in AI: GPT-6 Astra Launches, Open-Weight Leaderboard Reshuffles, Nvidia-Hugging Face Closes — September 4, 2026

September 4, 2026 — curated links and takeaways.

1. GPT-6 Astra Access Opens, Leads GPQA Diamond Reasoning Benchmark

OpenAI began opening access to GPT-6 Astra, its newest frontier model, with 96.0% on GPQA Diamond—the most discriminating reasoning benchmark. Developers evaluating long-context and reasoning workloads need fresh baseline comparisons; pricing and rate limits require immediate assessment against Claude and Gemini equivalents.

2. Kimi K3 Full Weights Released: 2.8T MoE Open-Weight Model

Moonshot AI released full weights for Kimi K3—a 2.8T-parameter Mixture-of-Experts model (50B active per token) under Modified MIT license on July 27, 2026. Open-weight deployment path now available for teams avoiding closed-model licensing; activation pattern enables cost-effective fine-tuning and distillation workflows.

3. WeatherNext 3: Google DeepMind Operationalizes Cyclone Forecasting AI

Google DeepMind released WeatherNext 3, targeting operational tropical cyclone forecasting with expanded coverage for underserved regions in Latin America, Africa, and Asia-Pacific. Weather AI models transition from research to production deployments; practitioners in climate-critical sectors can now deploy without legacy supercomputing costs.

4. LLaDA-Image: Fully Open Training Recipe for 6B Diffusion Transformer

Inclusion AI published LLaDA-Image, a 6-billion-parameter open-source Diffusion Transformer with training recipe, achieving state-of-the-art on Qwen-Image-Bench (53.53 EN) with 2–4 step inference turbo mode. Image generation now has reproducible, trainable open-weight baseline; builders can fork and optimize without licensing friction.

5. Nvidia Closes Hugging Face Acquisition for $12.9B, Commits to Open Ecosystem

Nvidia completed acquisition of Hugging Face for $12.9 billion, gaining control of a platform serving 18M developers and 200K companies, with formal commitment to support open-source and open-weight models. Consolidation risk: single GPU maker now controls primary distribution hub for open models; dependency on Nvidia's governance pledges intensifies.