Alibaba releases Qwen3.8-2.4T-A95B open-weight model
Alibaba released Qwen3.8-2.4T-A95B (Qwen3.8-Max), its largest open-weight model with 2.4T total parameters and 95B activated per token. It features a fine-grained MoE architecture with hybrid full and linear attention, a 1M-token context window, and up to 128K output tokens. On NVIDIA GB300 NVL72, it achieves over 4K tokens per second per GPU in FP8 precision.
How this story unfolded
4 days · 4 reports · 6 community posts · 10 of 15 shown
- Aug 12
- Aug 13
- Aug 14
- Aug 16
More stories today
Building Foundry: Weaviate turns archives into creative search
Foundry scans an existing archive read-only, records file facts, and syncs with Weaviate for keyword, semantic, or hybrid search. Test archive has 23 assets from five fictional projects with inconsistent names.
Weaviate Blog·2 hours ago
Scott Alexander explains mechanistic interpretability techniques
Scott Alexander's explainer covers the science of 'reading an AI's mind', noting modern AIs have millions of neurons and trillions of parameters. It highlights the 2023 breakthrough of many-to-many feature mappings, where combinations of neurons represent concepts.
Astral Codex Ten·2 hours ago
NVIDIA introduces CUDA Rust for native GPU kernels
NVIDIA announced CUDA Rust in September 2026, enabling GPU kernels written in Rust to compile natively to PTX. Two tracks are offered: SIMT and Tile, with Tile recommended first. NVIDIA plans to grow CUDA Rust into 2027 and beyond.
NVIDIA Developer Blog·2 hours ago
Qwen by email
Get an email when Qwen has news
No news that day, no email.
GPT-6 Astra rewrites Super Nintendo games in real time
Vaibhav Sisinty·2 hours ago
DeepSeek Flash 4.1 rolling out via API with new architecture
DeepSeek Flash 4.1 is in internal beta testing via API, featuring a new architecture with native multimodal support, stronger capabilities, faster speeds, and lower costs. Official release post expected soon.
Kimmonismus·3 hours agoOECD: Students Using AI Score Worse on Tests
The OECD's flagship educational report finds students who don't use AI chatbots outperform those who do, offering wide-ranging evidence the tech may hurt learning.
Bloomberg Technology·3 hours ago
AI health startup Forus raises $150M at $3B valuation
Healthcare automation startup Forus raised $150 million in a Series C round led by Bain Capital Ventures, valuing the company at $3 billion. The round comes just months after its previous funding.
Farmers embrace AI more than any other tech, McKinsey says
AI use is surging on farms as growers seek high-tech help in planning and managing operations after years of scaling back spending, according to McKinsey & Co.
