Daily AI Briefing
The 6 top stories of the 120 AIBriefs curated that day.
Muse Spark Open is an 816B-parameter mixture-of-experts model with 42B active parameters and a 1M-token context window, scoring 39 on the Artificial Analysis Intelligence Index — the #1 US open-weights model. Meta trained it on NVIDIA GB300 NVL72 for agentic coding, with multimodal perception across text, image, video, and audio.
Google DeepMind's Gemini 4 Argon scores 97.5% on ARC-AGI-3 and 91.3% on ARC-AGI-2, and tops Scale Labs' Humanity's Sixth Sense benchmark at 55.4%, edging GPT-6-Astra's 53.6%. It is available to developers, enterprises, scientists and consumers, and in Perplexity for Pro and Max subscribers.
Meta Superintelligence Labs released the Muse Wonder series, claiming SOTA on personal and professional agentic tasks; Wonder is on the Meta Model API and Muse Code, while Wonder Full is Muse Code subscribers only. Cursor lists Wonder at 46.1% on CursorBench, priced $4/M input and $20/M output.
DeepSeek calls the checkpoint a "Pareto improvement" over September's V4.1-Flash, beating it on all 11 reported benchmarks by 2.0 to 11.1 points. It scores 77.0 on DeepSWE v1.1 versus 74.0 for Claude Opus 5, but trails badly on HLE without tools (42.6 vs 56.3).
Opus 5.5 is priced at $4/M input and $20/M output tokens, with cache reads at $0.20/M — 20% cheaper per input/output token and 60% cheaper on cache reads than Opus 5. Anthropic estimates it costs about 40% less to run than Opus 5 for typical token-billed workloads, and says context per request has grown roughly 2.6x in six months.
OpenAI published 722 manuscripts (719 after three withdrawals lacking Lean proofs) across 372 families, including 90 of the top 500 open problems per Proof of Atlas. Results came from a single model averaging three hours of compute per solution after being asked to try ~4,000 problems.
Also that day
102 more stories that day — open the live feed.
Get tomorrow's AI brief in your inbox