Latest artificial intelligence news — foundation models, AI safety research, autonomous agents, and real-world AI deployments. Updated every 30 minutes from top AI publications.
Liquid AI's LFM3-1.7B, a 1.73-billion-parameter text model built on a tri-hybrid architecture of 13 gated short convolution layers, 5 sliding-window attention layers and 6 global attention layers across 24 total layers, …
Lightpanda, a headless browser written from scratch in Zig for AI agents and automation, reached version 1.0 on October 2, 2026, ending its beta after more than 10,000 commits, the company announced on its X account. The…
Telnyx has made OpenAI's open-weight gpt-oss-120b reasoning model available in beta on Telnyx Inference, hosted on Telnyx-owned GPUs in US regions. The model offers a 128K context window and configurable reasoning effort…
Microsoft released Mage-VL-Cowork, a 4B-parameter open-weight computer-use agent, on Hugging Face under Apache-2.0, with the company's model card reporting 66.0% on OSWorld-Verified under a 50-step budget across 361 task…
Six months after its launch, the Unitary Foundation's Clifft quantum-circuit simulator has passed 35,000 PyPI downloads and its paper has been cited 18 times, the foundation reported. Across 12 releases, Clifft added sym…
CopilotKit's OpenDuet, an MIT-licensed open-source proactive agent template built on the OpenBot platform, has drawn about 1,600 GitHub stars in the nine days since its October 21, 2026 release. OpenDuet reads Gmail, Cal…
A new arXiv paper (2610.30026) found that widely used document converters routinely discard highlighting and text colors, leaving AI agents unable to discern important instructions in enterprise documents. Of the 6,462 p…
Mintlify rebuilt its self-updating documentation architecture around a background agent and a runner agent sharing git-backed memory, raising its internal docs PR accept rate from 70% to 90%, software engineer Patrick Fo…
Unity Technologies shipped more than 15 Unity Industry releases in 2026, including the October experimental launch of Industry AI Workflows, which lets customers' own AI agents prep, optimize and publish CAD data across …
A Lean 4 repository by Isaac Rudich and Louis-Martin Rousseau of Polytechnique Montreal claims a deterministic Turing machine computes the smallest prime factor of any natural number N in N^o(1) time, with the repository…
A GPU fingerprinting technique using a standard PyTorch `torch.index_add_` call identified individual physically identical GPUs with 99.1% accuracy across 19 identical Nvidia A40s and 100% accuracy across seven identical…
The Vigil-SOC project released VigilSOC 1.0.0, the first generally available version of its open-source AI security operations center platform, following the 0.9.0, 0.10.0 and 0.10.1–0.10.3 releases. Version 1.0.0 harden…
The Qwen team and collaborators released Co-Design Bench, a benchmark of 273 real-user app-revision tasks with 2,331 hidden expert-written acceptance criteria drawn from 37,463 deduplicated queries across 26,314 coding s…
AMD released Early Access validated, optimized inference support for five MONAI Model Zoo bundles on AMD Instinct GPUs as part of its Life Science 26.08 medical-imaging stack. The five bundles — Vista3D, SwinUNETR (BTCV)…
Anthropic's Alignment Science team reported in October 2026 that a Claude red-team agent succeeded in every run at writing up to 5,000 fine-tuning examples that made a smaller research model misaligned while passing a tr…
Programmer Drew Breunig has built nesformer, a project that implements an original Nintendo Entertainment System's central processor inside the numerical machinery of an AI model, as documented on its GitHub repository a…
The cost of matching a landmark AI model's capability is now falling 11 to 25 times a year for levels set since OpenAI's o1 launched in late 2024, compared with 3 to 5 times a year for the GPT-4 and GPT-4o levels, accord…
Claude Code users reported that Anthropic began sending a server-side instruction around October 27, 2026 that makes the coding tool refuse work recreating existing video games, according to an X post by @chasmmmmmmmmmmm…
China's best open-weight AI models have outscored America's every day since September 2024, when Alibaba's Qwen2.5 overtook Meta's Llama 3.1 405B, according to Village's LLM Atlas, which scores 857 models on a single cap…
The best AI models from countries outside the US and China have trailed US models by a median of 9.9 months since 2024, according to Village's LLM Atlas, which scores 857 models on a single capability scale estimated fro…