Get tomorrow's AI brief in your inbox
Gemini Omni 1.2 Flash is generally available in the Gemini API and Gemini for business, generating native clips up to 15 seconds (up from 10) with multi-turn conversational editing. Pricing dropped 30% from Omni 1.1 to $0.07 per second for 720p video.
GPT-6 Cyber found 81 previously unknown high-severity Firefox flaws and helped build 948 Linux kernel exploits, scoring 48.5% on ExploitGym (up from 32.1% for GPT-5.6 Cyber). It went out of bounds under 1% of the time in testing, down from ~5%, and is gated to Daybreak Red customers.
18 of the 20 best open models today are Mixture-of-Experts. Dense is basically over. And yet the know-how to train a good MoE still lives mostly in the heads of a handful of teams. So we spent 800k GPU hours running the experiments ourselves and wrote everything down. Super https://t.co/7LphOrwoge
In a Masters of Scale Summit '26 conversation with Reid Hoffman, the Google DeepMind co-founder and chair argues frontier AI governance is lagging and that a more diverse group of experts needs a seat at the table.
AMD's data center revenue grew 97% in a single quarter, and Lisa Su says AI has turned once-a-year budgets into moving targets. She argues companies that set a budget annually and hold it are caught out by demand arriving faster than planned.
In a Sources Podcast interview, Mark Zuckerberg argues that the solution to AI safety problems is more AI rather than less. The episode is sponsored by Atlassian, Granola, and Mercury.
Intelligence Squared conversation with Mustafa Suleyman and Zanny Minton Beddoes on the AI revolution, asking whether AI has lived up to expectations and whether it will drive down costs.
A lawyer running DeepSeek-V4.1-Flash on eight 2018-era V100s reports 200+ tok/s on code and ~110 tok/s on legal writing, with 2,200 tok/s prefill at 128K context. All weights fit in VRAM with a 524K context window.
Walkthrough uses Meta's SAM3 model to track and identify any object in video in real time, pointing it at the rim, net and player to follow each across a clip. The agent is trained to locate each object and detect the moment of interest.
A Rust NYC talk by Bryan Cantrill examines how Rust fares in the LLM era. No transcript or detailed claims are available in the source beyond the video title.
JFrog predicted its fastest revenue growth in four years, citing AI giving hackers more sophisticated ways to test corporate defenses and driving demand for its digital security offerings.
After starting with simple copycats like Flappy Bird, Doodle Jump and Geometry Dash, the user says Opus 5.5 was good enough to produce 35+ games, then moved on to board games.
This is a Claude and Claude Code discussion subreddit to help you make a fully informed decision about using Claude and Claude Code to best effect for your own purposes. ¹⌉ Anthropic does not control or operate this subreddit or endorse views expressed here. ²⌉ If your problem requires Anthropic's help, visit https://support.anthropic.com/ This subreddit is not the right place to fix your account issues. ³⌉ For more help, check the resources below. ⁴⌉ Please read the rules before posting.
OpenAI customer story says Cult.fit teams across functions use Codex to build tools and speed up work. No metrics, pricing, or rollout details are disclosed.
Australian universities are widening ChatGPT access and embedding AI into teaching and learning to prepare students for an AI-enabled workforce, per OpenAI's own announcement.
Follow more of what you like — mute what you don't.
