Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Researchers introduced CompMat-Bench, a benchmark of 94 tasks derived from recently published computational materials studies, to evaluate AI agents on scientific research steps without rerunning expensive simulations. A…
A practical guide explains tokens per second (TPS) as a benchmark metric for measuring AI model processing speed, particularly for real-time NLP applications like chatbots and voice assistants. It outlines a methodology …
Runway announced Praxis-1, an open-weight world action model that converts its large-scale video pretraining into control policies for real robots, and said it is testing the model with early partners Noble Machines, Sta…
A developer released Sovereign-NB, a bare-metal x86-64 neural inference appliance written in Rust that boots as a UEFI application and serves fixed-size input frames over a UART stream. The runtime runs two integer-only …
Cloudflare released Clef and Clef-flash, its first models trained by the Workers AI team, as open-weight decision models under Apache 2.0 that return typed probabilities instead of free-form text. Clef is post-trained fr…
Cloudflare released Clef 27B, a 27 billion parameter multimodal decision model under the Apache 2.0 license that returns calibrated probabilities for a fixed set of answer options in a single forward pass rather than gen…
Cloudflare released Clef 27B, a 27 billion parameter multimodal decision model under the Apache 2.0 license that returns calibrated probabilities for a defined set of answers in a single forward pass instead of generatin…
Cloudflare's Clef 27B, a 27 billion parameter multimodal decision model released under an Apache 2.0 license, flagged a social engineering test with a 76% social engineering probability and only a 5.9% chance of granting…
Apple researchers Russ Webb, Amitis Shidani, Alice Bizeul and Dan Busbridge published an October 2026 paper showing that discrete diffusion samplers — including remasking and uniform-state samplers — only match the train…
Ian Maurer released ThinkThen 0.1, an open-source alpha library that wraps TypeSafe's Jev decision-model API — three hosted, zero-shot classifiers — in ten functions including decide, choose, tag, score, filter, rank, fi…
Google Cloud AI Research's RRSI paper reports that regularized recursive self-improvement of agent harnesses generalized to all six held-out benchmark splits, improving them by 1.8 to 4.7 points with no regressions, whil…
The Allen Institute for AI (Ai2) released Olmo-core 3 on October 1st, a redesigned open training stack for mixture-of-experts models that Ai2 says reached 52,000 tokens per second per GPU on a 47-billion-parameter MoE ac…
Carousel, a new homomorphic-encryption vector search system, scores an entire corpus under encryption and compresses results for client-side ranking, achieving recall@100 of 0.9715 across 1 billion BIGANN SIFT1B vectors …
David AI released DAI-ASR-I18N, a multilingual benchmark for conversational speech recognition and speaker diarization built from 147 hours of unscripted conversations across 21 languages. The company evaluated eight com…
Cohere released Embed 5, an embedding model family in two tiers — Embed 5 Pro and Embed 5 Fast — that share one embedding space so users can index with Pro and query with Fast, with both generally available on the Cohere…
Adaption Labs published research on October 1, 2026 evaluating its Invent API against APIs from Anthropic, Google, OpenAI, DeepSeek, and Zai on synthetic dataset generation in a zero-data regime. The technical report mea…
Amazon Payments applied a multi-objective contextual multi-armed bandit on Amazon SageMaker AI to personalize a product acquisition funnel, reporting a high single-digit percentage relative lift in final-funnel conversio…
Amazon Web Services released Strands Decider 2B, an open-source decision model inspired by TypeSafe's Jev, the same week OpenAI announced a similar offering. The model, built on the Qen3.5-2B LLM torso, sorts between pre…
Independent LLM-from-scratch experiments found that OpenAI's GPT-2 small (124M parameters) consistently outperformed the author's 163M-parameter models on an instruction fine-tuning task adapted from Sebastian Raschka's …
A team of researchers from Carnegie Mellon, MIT, New York University, and Stanford University built an AI called Ataraxos that beat Pim Niemeijer, arguably the best Stratego player of all time, 15 games to one with four …