Five at the frontier: Oct 13
Meta joins the model elite, Microsoft and Nvidia lay the racks for the next build-out, DeepMind proves theorems, and Washington blinks on safety testing.
Meta pushed into the top tier of AI models on October 12. The company released two models, Muse Wonder and Muse Wonder Full. The Decoder reports they score 56 on the Artificial Analysis Intelligence Index v4.3, up from 48 for their predecessor Muse Spark 1.3. TechCrunch, citing Meta's own testing, puts them level with Fable 5.1 and GPT-6 Astra, and The Verge says Meta claims they can go toe-to-toe with the industry's best. Muse Wonder runs through the Meta Model API. Muse Wonder Full is limited to Muse Code subscribers, according to TechCrunch. The gains cost something: The Decoder says the cost per task on the Intelligence Index climbed from $1.60 to $4.52.
The same day, Microsoft and Nvidia laid out the hardware for the next build-out. Microsoft said its third-generation Maia 300 chip will become the primary inference system for its products and services, deployed at multi-gigawatt scale from 2027, per The Verge and The Decoder. The Decoder reports Microsoft's claim of 40 percent more tokens per megawatt than leading GPU systems, and quotes chief executive Satya Nadella calling the chip 'our biggest bet yet on our own silicon.' Nvidia contributed its Vera Rubin rack designs to the Open Compute Project, with an updated specification that supports deployments of up to 250 kW.
Research landed alongside the products. A peer-reviewed study in Science, reported by The Decoder, found that Google DeepMind's AlphaProof Nexus solved nine open Erdős problems from a set of 353, two of them unsolved for 56 years. The system is built around Gemini 3.1 Pro and writes its proofs in Lean, a formal language that checks every draft. The Decoder contrasts DeepMind's choice of journal peer review with OpenAI, which it says faces a boycott for posting AI-generated proofs to GitHub without review.
Other work probed how far agents can improve themselves. Microsoft researchers found that clearing an agent's chat history while keeping its workspace improved results by up to 66 percent on the same compute budget, The Decoder reported. Their Fork-and-Flush method scored 0.78 on a normalised scale, against 0.47 for a single run. Sakana AI's MASS method, the outlet reported, let one model win 69.9 percent of comparisons against its base version after two rounds. There was a limit. When the model missed the same error across all its roles, it reinforced the mistake, something the authors call a potential 'Achilles' heel of recursive self-improvement.'
Policy moved too. The White House restored the UK AI Security Institute's early access to frontier models from Anthropic, Google DeepMind and OpenAI, The Decoder reported, reversing a September directive to stop sharing them. It attributes the turnaround to diplomatic pressure from UK AI minister Kanishka Narayan. Microsoft president Brad Smith, speaking on a BBC podcast, called for urgent rules, saying milk and washing machines have more oversight than AI companies, TechCrunch reported. He proposed approval from outside evaluators before products ship, and kill switches to pause systems humans can no longer control.
Two voices framed the stakes. Andrej Karpathy, an OpenAI co-founder who recently joined Anthropic, estimates about 5,000 people have access to internal frontier systems while roughly 6 billion have barely used large language models, The Decoder reported. And the technology keeps finding new ground. TechCrunch reported that WindBorne, which flies about 700 autonomous balloons steered by AI weather models, won a US Air Force contract worth up to $49 million.
Sources
- Meta's new frontier model Muse Wonder Full breaks into the AI top tier · The Decoder
- Meta releases its first frontier model, Muse Wonder · TechCrunch, AI
- Muse Wonder Puts Meta Back at the AI Frontier · Wired, AI
- Meta has some Wonder Full new AI models · The Verge, AI
- Microsoft plans to scale its own Maia AI chips to gigawatt levels by 2027 · The Decoder
- Microsoft’s new AI chip makes it less reliant on Nvidia’s GPUs · The Verge, AI
- OCP Global Summit: Amid Ecosystem Adoption, NVIDIA Opens Vera Rubin Architecture to Help Scale AI Factories Worldwide · NVIDIA Blog
- Google Deepmind's AlphaProof Nexus solved dozens of open math problems and actually got peer reviewed · The Decoder
- Clearing chat history helps AI agents find better solutions · The Decoder
- Sakana AI's self-improving model gets better at solving and judging, but can't catch its own blind spots · The Decoder
- White House reverses course and restores UK AI Safety Institute's access to frontier models · The Decoder
- Microsoft president says UK is best placed to drive AI regulation · TechCrunch, AI
- Anthropic's Andrej Karpathy says "the external world has seen the preview" of what AI can actually do · The Decoder
- WindBorne wins a hefty US Air Force contract for its balloon data · TechCrunch, AI