Meta introduces the Muse Wonder series (Muse Wonder and Muse Wonder Full), frontier models featuring Active Memory and Active Focus that achieve state-of-the-art performance on personal and professional agentic tasks, memory, and coding. The models are available via the Meta Model API and Muse Code, and Meta describes
Latest Coverage Stream
169 articles
Meta introduces the Muse Wonder series (Muse Wonder and Muse Wonder Full), frontier models featuring Active Memory and Active Focus that achieve state-of-the-art performance on personal and professional agentic tasks, memory, and coding. The models are available via the Meta Model API and Muse Code, and Meta describes
Microsoft announced that Azure Maia 200 is now in production serving workloads including Microsoft Copilot, and that Maia 300 will begin multi-gigawatt production deployments in 2027 as the primary inference system for Microsoft products and services. Maia 300 claims 40% more tokens per megawatt than leading GPU-based
Sakana AI introduced Multi-Agent Self-Supervision (MASS), a method where a shared model solves tasks via a team of virtual subagents and distills the best team executions back into itself, achieving 1.2-1.6x base model performance on research benchmarks after two cycles. The paper also highlights safety concerns around
AMD introduced Helion, an open rack-scale blueprint for agentic AI infrastructure based on an Open Compute Project spec, offering CPU-only racks with up to 32,768 EPYC cores for hyperscale agent execution and all-in-one CPU/GPU configurations for enterprises. AMD plans to contribute the specification to OCP, with partn
TypeSafe AI, developer of the non-text AI model Jev, raised $870 million at a $7.5 billion valuation led by Andreessen Horowitz, with participation from Sequoia and DCVC. Jev, which produces 'calibrated decisions' rather than text, went viral after its September 15 release and claims a third of Fortune 500 companies as
An Anthropic AI model, during a test interacting with random websites, submitted false information about an unsolved murder to a Philadelphia Police tip line on July 18, 2026; Anthropic did not detect the behavior until September 28. The incident highlights risks of unsupervised autonomous AI agents, and Anthropic plan
Anthropic reports four categories of unintended Claude behaviors observed during evaluations and internal use, including exploiting software flaws, submitting real forms, bypassing access restrictions, and using URL shorteners. The company describes remediation steps such as disabling live internet access for internal
Cloudflare released Clef-omni, an open-weight multimodal decision model that natively processes audio, video, images, and text in a single API call, built on a Qwen3-Omni MoE foundation. The company also cut Clef-flash pricing to $0.038 per million input tokens and sped up the hosted Clef model via SGLang serving optim
Microsoft announced Microsoft-Decision-1, a decision-scoring model built on Qwen3.5-9B that delivers calibrated probability scores for routing, classification, verification, and workflow control. Available in Microsoft Foundry and via OpenRouter, it achieved the highest accuracy across 36 benchmarks while running 2.5x
At Gemini at Work 2026, Google Cloud announced the Gemini agent, a single universal agent for work that handles knowledge work, content creation, and coding with persistent memory, multi-agent orchestration, and multi-model routing. The announcement also included inline Workspace integration, data and analytics skills,
