ATLAS uses 547 tasks with a $0.24 grading cost to reveal that no agent achieves a row F1 score above 0.5 under a $1 budget.
Speculative decoding accelerates LLM inference by increasing total FLOPs to utilize idle compute resources, shifting the workload from memory-bound to compute-bound at low batch sizes.
Epoch AI's evaluation of six frontier models reveals that while closed-weight models like Claude Fable 5.1 and GPT-6 Astra excel at defined coding and computational tasks, they lack the judgment for open-ended work, often misinterpreting experimental flaws and failing to adhere to implicit standards.
The startup, valued at $285 million, is backed by Benchmark and Index Ventures and already counts Cognition and Modal as early customers.
The ML Intern agent trained, evaluated, and published six specialized models, including a 0.8B prompt rewriter and a 4-step text-to-image distillation, on Hugging Face hardware for about USD 103.
The lightweight model generates 1-15 second video clips with native audio in resolutions up to 1080p.
Human participants achieve 93.1% accuracy on the new HSS benchmark, while the top-performing model, GPT-6-astra, reaches only 53.6%.
The new platform offers token-based billing with $5 in free credits for new accounts and supports synchronous parsing for documents up to 50 pages.
The Long Term Benefit Trust holds four of seven board seats and controls key governance mechanisms, though its trust agreement remains non-public.
Perplexity has released pplx-embed-v2-late, a family of ColBERT-style multi-vector embedding models in 0.6B and 9B sizes, supporting multimodal retrieval for text and visual documents.
The new configuration option reduces Fast Origin Transfer usage and improves time to first byte by preventing request bodies from being sent to Routing Middleware.
Snapshot restoration time now counts toward creation, adding about 50 ms to p50 latency but eliminating wait times for the first operation.
StepFun's flagship model supports text and image input with a 1M-token context window for agentic coding and analysis.
OpenAI disrupted two sophisticated influence operations using AI to create fake personas and launder geopolitical narratives, marking the first Category 5 breakout in its reporting history.
Apple researchers introduced Normalizing Trajectory Models (NTM), a framework that matches strong baselines in four sampling steps while retaining exact likelihood over the generative trajectory.
Google Cloud reports that Irish organizations, from Ryanair to startups, are deploying Gemini Enterprise to advance technical foundations and workflows.
Google announced the Gemini agent, a universal AI for work that operates across Google Workspace and Microsoft 365, featuring multi-agent orchestration and persistent cloud execution.
It proposes a multi-layered defense strategy at the system, model, and user levels based on contextual security theory.
It supports isolated multi-user agents within a single process and provides 40 LLM presets and an expert template library.
The new MA-BC method pools demonstrations only when expert actions agree, providing upper and lower bounds on sample complexity.
Broadcom is arranging more than $50 billion in financing for OpenAI's custom AI chip.
The dataset covers 106 SLMs with 150M parameters or fewer, evaluated on a Runpod B200 under Apache 2.0 license.
Sarah Yager, OpenAI's first human rights and responsible deployment lead, expressed deep concern about military targeting and autonomous weapons during a conference appearance.
The new models include a 29B Pro version and a 3B Lite version, with immediate support for ComfyUI and Diffusers.
AI Briefing