Mistral’s new 1T model aims to leapfrog closed and open rivals
The frontier moves again: Mistral's new trillion-parameter model — making the rounds as Mistral Large 4, topping Hacker News at 1,253 points — aims to leapfrog both closed and open rivals in one shot.
LibreOffice says ‘no AI’ is now a software feature
The backlash becomes a bullet point: LibreOffice is advertising the absence of AI as a feature. When 'does not phone home to a model' is a selling point, the feature debate has officially inverted.
The next hurdle for AI agents: getting websites to let them in
The agent era has a bouncer problem: the next hurdle is getting websites to let agents in at all — permissions, bot defenses and terms of service are the new frontier for autonomous software.
Beam, day two: 501B total, 23B active, weights later this month
Reflection's Beam is a sparse mixture-of-experts model — 501 billion parameters, 23 billion active — trained on 23.8 trillion tokens with a 100-million-rollout RL run on 10.5K GB300 GPUs. The company positions it near GLM 5.2 and approaching Qwen 3.8-Max on coding and agentic tasks, with Kimi K3 still ahead on raw capability — Beam's pitch is inference efficiency, not benchmark crowns. Caveat: the weights, technical report and model card land later this month; today it's an announcement, not a download.
Anthropic gives startups a free year of Claude Team plus $1,000 in credits
The platform war, subsidies edition: Anthropic is offering startups a free year of Claude Team and $1,000 in API credits — locking in the next generation of AI-native companies early.
Hark releases an AI personal assistant with a focus on privacy
Another entrant in the private-assistant race: Hark's AI personal assistant leads with privacy as its headline feature — a pitch that's becoming a category of its own.