OpenAI says it stopped a coordinated campaign in which more than 15,000 accounts tried to copy the hidden reasoning of its models. The company ties part of the activity to people connected to Moonshot AI. But researchers found that the same attack kept working on Microsoft Azure for weeks, even against the new GPT-6 Astra.
The AI system Ataraxos has decisively beaten the best Stratego player of all time. The board game is a tough test for AI because both sides set up their pieces face down. Google Deepmind fell short in 2023 despite a multimillion-dollar budget. Researchers from Carnegie Mellon, NYU, Stanford, and MIT built Ataraxos for less than $8,000.
... but you can’t try it yet unless you are “government users and trusted cyber defenders in the Fairwind Program”
Gemini 4 Argon is Google's first frontier model in over seven months. It matches GPT-6 Astra in independent testing but can't keep up with Anthropic's Claude Opus 5.5. The per-token price is low, but Argon burns through more than twice as many tokens per task as Astra. Select testers get access first, with the API and paid tiers following later.
Gemini 4 Argon delivers frontier performance in complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense. Today, we’re announcing our new frontier model, Gemini 4 Argon, which is rolling out to a set of trusted cyber defenders through our Fairwind Program.
Liquid AI released experimental draft model LFM2.5-VL-DSpark for vision-language model LFM2.5-VL-3B. Speculative decoding speeds it up without changing output quality: decoding improves by up to 3.13x on devices and 2.66x on H100, with end-to-end gains of up to 2.62x and 2.27x respectively.
Last week TypeSafe AI unveiled Jev , their first example of a new category of model that they are calling "System One models" (I'm with Maggie Appleton, I think "decision models" is a better name for these).
The definitive Jev podcast with its lead creator.
System performance, efficient infrastructure scaling and continuous software optimization are key levers that determine AI inference economics. Higher system performance means more tokens generated, resulting in higher revenue. Efficient scaling means throughput grows proportionally as hardware gets added, requiring fewer resources to serve users at scale. Continuous optimization means generating more value from infrastructure investments.
TypeSafe released Jev, a non-autoregressive model trained with RLCD solely for decisions, classification, routing and scoring. Official claims put it at 20–200 times faster and 40–400 times cheaper than small frontier LLMs, with no output token charges.
Google Research proposed Retrieve-for-Train, using offline reinforcement learning to find reward-aligned query fan-out and compile it into supervision, then distilling it into a 53.9M-parameter diffusion retriever. At inference, it performs one non-autoregressive query fan-out without generating CoT reasoning tokens.
Ian Buck, vice president of hyperscale and high-performance computing at NVIDIA, Tuesday spoke on AI factory efficiency at the AI Infra Summit, the Santa Clara Convention Center event that has morphed into a Coachella of infrastructure tech.
We’re sharing an AI-generated solution to the Navier–Stokes Millennium Prize Problem, including a writeup and a formal proof in Lean.
A Hugging Face blogger reproduced Surya Narreddi's idea of having a language model paint watercolours using TRL and OpenEnv. The model paints through around 150 lines of JavaScript with p5.brush, with datasets, RL environments, training scripts and models all open-source.
Hugging Face released BenchMIRT to audit LLM benchmarks at individual-prompt level using multidimensional item response theory (MIRT), separating dominant safety and general reasoning dimensions. Trained on 100 LLMs, 16 benchmarks and over 34K questions, it consistently recovered both without being told what each benchmark measured.
The IBM Granite team released Granite 4.2, its first dense, decoder-only reasoning model family, in 3B, 8B and 30B sizes, all open-source under Apache 2.0.
Multiverse Computing published a paper proposing Quantization-Aware Healing (QAH). After compressing GPT-OSS 120B to 60B parameters and quantising it to MXFP4, the method distils directly from the original uncompressed model rather than a reconstructed bfloat16 checkpoint.
Liquid AI released DSpark draft model checkpoints for LFM2.5-1.2B-Instruct, LFM2.5-2.6B and LFM2.5-8B-A1B. Speculative decoding accelerates decoding without changing output quality, increasing throughput by up to 3.18 times on GPUs and 2.87 times on devices.
The new frontier of AI is developing capable autonomous researchers
A path, a fence, a knot. MindTopo sets a new benchmark for testing how AI understands topological relationships and highlights new opportunities to strengthen spatial reasoning and planning.
Hugging Face's ALTK-Evolve and ACE both provide agent memory without weight updates, but differ in delivery: ACE injects the full playbook at every step, while ALTK-Evolve retrieves a small number of strongly supported guidelines for each task.
Google Research released the Chain-of-Evidence (CoE) verifiability framework, implemented in a Science One Framework prototype, with automated CoE Audit metrics.
Figure 1: CUDA-to-MLX optimization translation map. CUDA optimization knowledge can be translated into architecture-native MLX strategies rather than copied instruction-for-instruction. We face a new epoch in computing. Hardware is changing rapidly — not just faster GPUs, but a growing range of chips from different vendors, each with its own architecture and often tailored to specific AI workloads.
Overview of ABBEL compared to traditional recursive summarization. Beliefs replace the full interaction history as the agent’s working context, and belief grading improves performance by supervising the contents of each belief state.. As task horizons grow, LLM contexts can’t scale forever.
Thinking Machines released Inkling on Hugging Face, a multimodal MoE model with around one trillion parameters, a one-million-token context and native image, text and audio inputs, alongside Inkling-Small with 276 billion total and 12 billion active parameters.
The third Hugging Face PyTorch profiling instalment traces attention implementation with a profiler. Comparing naive attention with in-place masking shows replacing masked_fill with masked_fill_ removes one GPU Memcpy kernel per forward pass.
Import AI 464 reports Fable wrote what maintainers called the first genuine and fastest megakernel on KernelBench-Mega. CUDA on RTX PRO 6000 Blackwell achieved 18.71x acceleration, versus 14.4x for Claude Opus 4.8, 11.14x for GLM-5.2 and 4.34x for GPT 5.5.
Mistral released Leanstral 1.5 under Apache-2.0, with 119B total and 6B active parameters and major formal verification improvements. It achieves 100% on miniF2F, solves 587 of 672 PutnamBench problems, and sets current best results of 87% on FATE-H and 34% on FATE-X.
Google Research published an architecture attaching multi-token prediction (MTP) heads to frozen Gemini Nano v3 models, accelerating on-device inference without changing backbone weights. It has shipped with the Pixel 9 and 10 series.
A Google Research paper at COLM 2026 finds reasoning traces unlock factual knowledge LLMs otherwise cannot recall, even for simple single-hop questions. Tests on Gemini-2.5 Flash and Pro and Qwen3-32B identify two mechanisms: extra tokens act as a 'computation buffer', and 'factual priming' produces related facts to semantically prepare the correct answer. Self-generated intermediate facts can also introduce hallucination risks.
Google DeepMind released experimental open-source model DiffusionGemma, generating text blocks in parallel through text diffusion and achieving up to 4 times faster inference on dedicated GPUs.
When will markets price the singularity?
Overview of adaptive parallel reasoning. What if a reasoning model could decide for itself when to decompose and parallelize independent subtasks, how many concurrent threads to spawn, and how to coordinate them based on the problem at hand? We provide a detailed analysis of recent progress in the field of parallel reasoning, especially Adaptive Parallel Reasoning.
GRASP is a new gradient-based planner for learned dynamics (a “world model”) that makes long-horizon planning practical by (1) lifting the trajectory into virtual states so optimization is parallel across time, (2) adding stochasticity directly to the state iterates for exploration, and (3) reshaping gradients so actions get clean signals while we avoid brittle “state-input” gradients through high-dimensional vision models.
Google proposed Simula, a reasoning-first synthetic data framework reframing dataset construction as mechanism design. It builds datasets from scratch without seed data through global diversification, local diversification, complexification and dual-critic quality checks.
Google Research released TurboQuant, compressing the KV cache to three bits without loss of model accuracy or training and fine-tuning, alongside the QJL and PolarQuant methods.
Mistral AI released Mistral Small 4, the next major version in the series and its first model to combine Magistral reasoning, Pixtral multimodal capabilities and Devstral agentic coding in one model, under Apache 2.0.
Mistral AI released Leanstral, the first open-source coding agent for Lean 4, using a sparse architecture with 6B active parameters. Its weights are available under Apache 2.0, and it is integrated into Mistral vibe and the free labs-leanstral-2603 API endpoint.
--> Understanding the behavior of complex machine learning systems, particularly Large Language Models (LLMs), is a critical challenge in modern artificial intelligence. Interpretability research aims to make the decision-making process more transparent to model builders and impacted humans, a step toward safer and more trustworthy AI.
In this post, I’ll introduce a reinforcement learning (RL) algorithm based on an “alternative” paradigm: divide and conquer . Unlike traditional methods, this algorithm is not based on temporal difference (TD) learning (which has scalability challenges ), and scales well to long-horizon tasks. We can do Reinforcement Learning (RL) based on divide and conquer, instead of temporal difference (TD) learning.