Mistral has released the weights of its 1-trillion-parameter Large 4 under a modified MIT licence. Here is what you get, the hosting catch, Mistral's benchmark claims and what it takes to run it.
Google's new Gemini 4 Carbon is aimed at finding and fixing software flaws, but only vetted cyber defenders can use it for now. Here is what Google claims, who gets access, and what is unverified.
OpenAI says its AI disproved the 27-year-old Exponential Time Hypothesis and that the result checks out in Lean. Experts at UC San Diego and MIT are now verifying it. Here is what it means and what is unproven.
OpenAI's new GPT-6.1 Terra costs $1 per million input tokens and claims to match its flagship on one benchmark at a tenth of the price. Here are the claims, the pricing and a reality check.
Reports say OpenAI is discounting heavily to win Anthropic's biggest customers. The 90% headline is one customer's cost cut, not a price list. Here is what is verified and what developers should do.
Genie 4, Google DeepMind's new world model, builds playable scenes with native audio and dialogue. Here is what it does, who can try it and its known limits.
OpenAI's frontier models and Codex workflows are heading to Google Cloud's Model Garden, bought through Marketplace. It is a private preview for now, with general availability promised in the coming weeks.
Granite Speech 5.0 is IBM's free, open speech model: transcription in six languages, fast on-device recognition and better meeting notes. What it does, and lacks.
Phonon TTS is a free text-to-speech model from Paris startup Gradium that runs offline on phones, laptops and a Raspberry Pi. What it does and how to try it.
Block's Dessa speech model learned to talk from sound alone, with no text. Here is what the research preview does, who built it and who can use it now.
Kimi K3.1 Pro scores 65.7% on Terminal-Bench 4.0, beating GPT-6 Astra on Moonshot's numbers. Here is how to use it now and what the open weights mean.
The Agentic AI Foundation takes in four projects from Anthropic, OpenAI, Hugging Face and Alibaba. Here is what each does and why it matters if you build agents.
Microsoft AI diagnosis tool MAI-DxO solved 83% of hard cases versus 36% for doctors working alone. Here is how it works, the catches, and what comes next.
The OpenAI Decisions API is now open to all developers. It answers bounded questions up to 10x faster than GPT-6 Luna and charges nothing for output tokens.
Canva Craft is Canva's first in-house image model: 2K images, 30 aspect ratios, readable text and layered edits, free in Dream Lab and via the Canva API.
Get a notification when something new is published.
Done. To stop them later, use the notification settings for this site in your browser.
