The Day in AI

• Microsoft announced Maia 300, its third-generation AI chip on TSMC N2P, and says production rollout at multi-gigawatt scale starts in 2027.
• Meta launched Muse Wonder and Muse Wonder Full, and Digital Trends reports the Full model ranks second on the Artificial Analysis Intelligence Index.
• Anthropic reports that a 9B model got inference running on 8 of 9 hardware types in its self-replication experiments.
• Zhipu began a limited test of GLM-5.3 HighSpeed in ZCode, where one tester saw output above 490 tokens per second.
• DeepSeek reportedly led a seed round for a robotics AI startup founded by a former Tencent scientist.

Maia 300 follows Maia 200, which Microsoft announced in January. It uses a chiplet design that puts compute, the memory controller and I/O on separate dies, and carries 288GB of HBM4 in eight stacks. A full rack holds 96 accelerators and draws about 167 kW, according to Reuters. Microsoft reports up to seven times the rack-scale token throughput of Maia 200, and Scott Guthrie expects 40% more tokens per megawatt than leading GPU-based systems. Microsoft says it will scale Azure Maia to multi-gigawatt deployments starting in 2027.

Meta launched Muse Wonder and Muse Wonder Full, which it calls its largest and most capable models so far. Both add Active Memory, which keeps user preferences across chats, and Active Focus, a loop that tries other approaches when a task fails. Digital Trends reports that Muse Wonder Full ranks second on the Artificial Analysis Intelligence Index behind Claude Opus 5.5 and ahead of GPT-6 Astra and Gemini 4 Argon, and one third-party article puts it six points behind Opus 5.5. Muse Wonder is in the Meta Model API and Muse Code, and Muse Wonder Full is for Muse Code subscribers. Muse app plans cost $20 or $100 per month and Muse Code plans $50 or $200 per month. One report puts API input at $4 per million tokens, the same as Claude Opus 5.5.

Anthropic's Frontier Red Team published research arguing that small models may be months away from copying themselves onto commercial IoT devices such as smart ovens and off-brand Android TV boxes. In its tests, models of 30B parameters and under could complete many of the individual replication steps but made mistakes when chaining them together. A 9B model got inference running on 8 of 9 hardware types, according to Anthropic. A replicator like this would have no central point to shut off and could work out unfamiliar devices by itself, which ordinary malware cannot. Anthropic expects it would spread slowly, with a doubling time of days.

Zhipu started a gray-release test of GLM-5.3 HighSpeed in its ZCode coding tool. It keeps the 1 million token context and 131,072 token maximum output of GLM-5.3 and differs only in speed. One tester saw output above 490 tokens per second, against 70.84 for the standard GLM-5.3 API measured by AI Ping, and that figure from a small test may not hold at scale. An earlier GLM-5.1 high-speed API reached 400 tokens per second with the TileRT team, which compiles the inference graph ahead of time so it does not change at runtime.

• Z.ai announced GLM-Legal, a legal agent model built on GLM-5.3-Flash, and reports a 19.2% all-pass rate on Harvey LAB.
• NVIDIA contributed Vera Rubin rack designs to the Open Compute Project through MGX 1.2, which supports racks up to 250kW.
• Runway showed Project Continuum, a demo operating system where parts of the screen are real-time generated video instead of apps.
• StepFun released Step 5 Preview on 8 October, a 600B-total, 27B-active mixture-of-experts model at $1 per million input tokens and $2.70 per million output tokens.
• DeepSeek reportedly led a seed round for a former Tencent scientist's robotics lab, with reports differing on whether $400 million is the round size or the valuation.
• Huawei Noah's Ark Lab reported that its Memento 3 agent cleared all 25 ARC-AGI-3 public games with a frozen model, according to Pandaily.
• Mistral AI unveiled ML4, an open-weight model, according to Digitimes.
• OpenAI launched GPT 5.6 Cyber and expanded its Daybreak security service, according to Digit.