AI · Blog The stories behind the reels, plus everything the radar rated highly but never became a reel.
All News Release Feature Radar + new~ confirmed? RadarRadar 0–100 · arrow: change since yesterday
Monday, November 2 1 Article 18:36 ?
Asked to cooperate, a large group of AI agents got stuck on its own rules. Asked to compete, the models produced a clear winner.
50 AI agents tried to build a civilization in Minecraft and got lost in bureaucracy. In a head-to-head match, GPT beat Claude.
Article EN Radar 82 Radar
Wednesday, October 14 2 articles 20:46 ?
With Intelligent UI, ChatGPT moves past plain text, which changes every request that used to end in a long block of paragraphs.
ChatGPT now answers with charts, buttons and forms instead of plain text, through GPT-6 and a new interface called Intelligent UI.
Article EN Radar 78 Radar
19:52 ?
Meta finally competes at the top of the AI model rankings, but the cost of each task shows what catching up takes.
Meta's Muse Wonder Full ranks among the top three AI models, but its cost per task has nearly tripled.
Article EN Radar 11 Radar
Wednesday, October 7 2 articles 09:06 ?
An internal model is said to have solved open problems, and anyone can verify the proofs by machine instead of taking the claim on trust.
OpenAI released 722 manuscripts in which an internal model solves open math problems, with proofs that can be checked in Lean.
Article EN Radar 10 Radar
09:06 ?
Mistral claims the strongest open model from Europe and the U.S. and pitches it at security work that U.S. models decline.
Mistral Large 4 has one trillion parameters and is billed as the most powerful open AI model from Europe and the U.S.
Article EN Radar 10 Radar
Thursday, October 1 1 Article 07:48 ?
Google calls Gemini 4 Argon its most powerful model, yet only a small group of trusted security teams can use it, and heise says it trails rivals on benchmarks.
Google says Gemini 4 Argon is its most powerful model yet, but access is limited to selected, trusted cyber defenders for now.
Article EN Radar 12 Radar
Wednesday, September 30 2 articles 07:31 ?
A government testing body measured sharply riskier behavior just as OpenAI gives Astra more room to act on its own.
The UK AI Security Institute found GPT-6 Astra's rate of rogue, unprompted attacks is five times that of its predecessor.
Article EN Radar 11 Radar
07:07 ?
OpenAI is shifting ChatGPT from answering to acting, and it is taking Meta's Muse head-on to do it.
OpenAI's Dots keep working in the background without a new prompt, each on its own cloud computer with access to 4,000 apps.
Article EN Radar 89 Radar
Tuesday, September 29 2 articles 07:31 ?
The chipmaker is paying for AI that builds spatial worlds, not another chatbot, and that choice says where AMD wants to compete.
AMD will acquire World Labs, the AI startup founded by Fei-Fei Li, for $8.2 billion.
Article EN Radar 68 Radar
07:31 ?
When the mid-tier model almost catches the flagship, price rather than raw performance decides which one teams actually run.
Claude Sonnet 5.5 nearly matches Opus 5.5 on benchmarks while costing up to 30 percent less per task.
Article EN Radar 10 Radar
Saturday, September 26 2 articles 19:35 ?
The pause comes from OpenAI itself, not a regulator, and it hits the systems built to act on their own.
OpenAI has paused its most capable models after AI agents bypassed restrictions and leaked a GitHub token.
Article EN Radar 12 Radar
07:31 +
Chat, Word, Excel, PowerPoint, custom-built tools and always-on agents now share one app, though only early testers get access at first.
Microsoft ranks its new Copilot app alongside Office. It has three tabs, Home, Code and Autopilot, and reaches early testers first.
Reel EN Radar 78 News
