Frontier Labs
model releases, research and announcements from OpenAI, Anthropic, Google, Meta, xAI, Mistral and the labs building frontier AI
FLUX.3 Image Klein Lets You Draw a Box and Tell the Model Exactly Where to Put Things
Black Forest Labs' new open image model combines text-to-image and multi-reference editing with bounding-box layout control in a single checkpoint, generating a 1024-pixel image in about two seconds on a high-end GPU.
Falcon-Emirati-7B Can Actually Speak Emirati Arabic, Not Just Translate It
The Technology Innovation Institute has released a 7-billion-parameter model trained specifically on the Emirati dialect, scoring 84.83% on a purpose-built benchmark where models far larger fall short.
OpenAI Safety Writer Quits and Calls Out the Company Publicly
David Robinson helped write the safety reports for every major OpenAI model release. Now he's gone, and he wants the public to know why.
The Allen Institute Just Open-Sourced a Training System for Trillion-Parameter AI Models
Olmo-core 3 lets academic labs and smaller teams build huge AI models without the usual sky-high compute bill. Here's what changed and why it matters.
Apple Asks Whether AI Coding Agents Actually Need All That Extra Machinery
New research from Apple ML Research finds that a single, well-trained AI agent given direct access to a computer can match or beat far more complex multi-agent systems on real machine learning tasks.
Google's new frontier model can write a million tokens without stopping
Gemini 4 Argon targets software engineering and cybersecurity, but its phased rollout tells you something about how cautious Google is feeling.
OpenAI Coverage Fell 83% in Our Feed Last Week. Here Is What That Actually Tells Us.
AI2Day published 9 OpenAI stories in the week to 28 September, down from 52 the week before. The story shape changed too.
OpenAI Is Asking Elite Mathematicians to Tell It What It Got Wrong
After a string of maths results that damaged its reputation, OpenAI has formed a panel of independent mathematicians to advise the industry on where AI and academic research should and should not mix.
Google gave its voice AI a face that talks back at you
Gemini 3.8 Live with Live Avatar is generally available in Gemini Enterprise as of September 24, 2026. It lip-syncs across 97 languages, fetches data mid-sentence, and stamps every frame with an invisible watermark.
Liquid AI's vision model gets a speed boost of up to 3.13 times with no change to what it says
A 280-million-parameter add-on lets LFM2.5-VL-3B generate text up to 3.13 times faster on Apple silicon, leaving output word-for-word identical.
Sam Altman rules out an OpenAI IPO this year and says uncontrollable AI is possible
The OpenAI chief told Fortune that 2026 is the wrong moment to go public, and admitted that an AI system beyond human control could theoretically be built. He promised to halt development before that happens.
OpenAI Built Its First AI Chip Using Its Own AI, And Did It in Under 20 Months
The Jalapeño chip promises to cut response times by up to 3.6 times compared to the Nvidia hardware OpenAI currently relies on. The wilder story is how a team of fewer than 100 people built it so fast.
Google's new Gemini voice models can think and talk at the same time
Two fresh Gemini Live models promise faster, smarter voice chats and voice agents that finish tasks while they keep talking.
Apple researchers taught an AI coding assistant to remember what actually worked last time
A new memory system from Apple ML Research lets AI coding agents carry useful knowledge from one work session to the next, rather than starting from scratch every time.
25 Fields Medal Winners Have Signed a Letter Challenging OpenAI's Maths Claim
OpenAI says its AI solved a century-old maths problem. The world's top mathematicians say the company needs to be honest about what that actually means.