AI — Page 6

The preview injects a WebMCP bridge at Cloudflare's edge, giving browser agents structured access without origin-side deployment.

Meta reported perfect physics theory scores, while leaving the models and evaluation setup unspecified.

The account behind the claim calls Astra OpenAI's largest pretrain since GPT-4.5 and names "mewfour" as its release candidate.

The release includes WeatherNext 2 and WeatherNext Cyclones checkpoints, plus a lighter version for researchers and weather agencies with constrained compute.

The public beta doubles Wan's prior 15-second ceiling and starts at 5 cents per generated second through Alibaba Cloud and Qwen Cloud.

The multiyear contract is worth at least half Mirendil's seed financing, although the frontier AI lab has yet to release a public product, benchmark or customer deployment.

Edo Liberty and Ash Ashutosh are expanding Pinecone beyond vector storage with a governed knowledge layer that runs in the customer's cloud.

This one isn’t a blowout, but DeepSeek-V4-Flash finishes ahead by being more dependable on code, bug-fixing, and instruction-heavy writing tasks. Cohere-command-a-plus-05-2026 lands some sharper wins on SQL correctness and extraction, yet not enough to erase DeepSeek’s broader edge.

Google will invest in and provide cloud services to Discovery Loop, while Koray Kavukcuoglu takes over Google DeepMind's daily operations from Demis Hassabis.

Liang Wenfeng's lab has not published new rates or an effective date, leaving developers to price in an undefined increase.

Irregular founders Dan Lahav and Omer Nevo discovered the incident while testing Meta's model, exposing the risks of realistic AI-security evaluations.

Apple GPU architecture alumnus Jacob Peake is targeting frontier labs with sealed chip-design tasks and physics-based grading.

This one wasn’t close. Across four very different prompts, Flux 3 Text To Video Draft consistently followed the brief more faithfully, while Heygen Video Agent repeatedly delivered cleaner-looking but less correct videos.

The Drift co-founder will run Composer and Customer Agent after the Q3 deal closes, reuniting with Klaviyo co-founder Andrew Bialecki.

One model brought occasional flashes of discipline; the other delivered far more complete prompt execution. This matchup turns on consistency across text, counting, layout, and scene-specific details rather than on isolated aesthetic wins.

A Neon case study says the specialized open-weight model matched GPT-5.6 Sol on retrieval accuracy while costing about 100x less per inference workload.

Mark Zuckerberg's terminal agent uses persistent subagents and Muse Spark 1.2 to plan, edit and validate code.

Controlled experiments and a July containment failure show how tool-using models can turn ordinary software flaws into autonomous attack chains.

YC S26 founder Yahia Bakour says the four-person company serves 400-plus customers after expanding beyond its Brand.dev roots.

The Apache-licensed release includes weights, post-training code and deployment tools, while Xiaomi keeps the main training corpus private.

Brett Adcock's self-funded AI lab beat the previous benchmark leader by 0.4 points, while broader performance claims rely on Hark-run tests.

The proposed reverse acquihire would give Google AI training technology and part of the 35-person startup's team without buying Mechanize outright.

Google will take an equity stake and provide first-year compute as the founders use Discovery Loop to improve machine-learning systems.

Cloudflare's AI Gateway ties verified identities to model spending, while User Insights compares session costs with a 30-day baseline.