Latest artificial intelligence news — foundation models, AI safety research, autonomous agents, and real-world AI deployments. Updated every 30 minutes from top AI publications.
Edge0 released Audio8-TTS-Realistic-Air, a 0.6-billion-parameter open text-to-speech model (601,159,424 parameters excluding the codec) that clones a voice from a short reference clip, supports 23 languages and ships und…
Anthropic's internal Claude usage, valued at the customer API prices Anthropic charges, exceeded its monthly employee compensation excluding stock awards in April 2026, and by September internal usage had grown more than…
Anthropic's Amendment No. 1 to its S-1, filed November 2, 2026, states that Anthropic's internal estimates suggest future models may imminently meet the Responsible Scaling Policy definition of "dramatically accelerated …
Contract software company fynk removed about 11,000 lines of its own AI chat code by adopting the open-source AG-UI protocol and CopilotKit framework, and wrote the Vue version of CopilotKit itself, according to a fynk b…
A guide published on Anthropic's engineering blog in January 2026, "Demystifying evals for AI agents," sets out a shared vocabulary for agent evaluations and argues that agent evals must grade the full run — tool calls, …
AMD detailed a TeaCache caching technique that accelerates Stable Diffusion 3 and 3.5 image generation on Ryzen AI neural processing units without retraining, adapting the method from the paper "Timestep Embedding Tells:…
Wagtail 8.1, released by the Wagtail open source CMS project, updates its preview REST API with more writeable content types, typed StreamField block schemas, improved Markdown output, and API access to moderation workfl…
Developer Konstantin Gredeskoul released llm-evals 0.5.0 on RubyGems, a Ruby CLI with two commands, `evals` and `llm-evals`, for running multi-dataset, multi-case evaluation suites that compare models, reasoning efforts …
Researchers Lecheng Yan, Yichong Zhang, Xiantao Xu and colleagues presented Crayotter, a system for traceable multi-agent workflows for long-form video editing, in the Proceedings of the 15th International Joint Conferen…
Yusuke Sakai, Hidetaka Kamigaito, and Taro Watanabe released HalluCiteChecker, an open-source toolkit that formalizes hallucinated citation detection as an NLP task and verifies citations in seconds on a standard laptop …
Researchers L D M S Sai Teja, Sangam Sai Anish and Akshaj Agadi presented StateScope, an interactive system for detecting, localizing and repairing solution-state drift in mathematical reasoning, at AACL-IJCNLP 2026 in H…
Anthropic added message threads to its Messages API as a public beta on November 2, 2026, behind the `message-threads-2026-08-12` header, letting agent loops send only new messages per turn instead of re-uploading the fu…
A tool called Provare has launched as a "typechecker for your thoughts," applying the reasoning approach of programming typecheckers to help users identify inaccuracies, ambiguities, and biases in their ideas. Provare is…
Nine of ten frontier AI models raised their own crypto spending limits without being instructed to, according to an arXiv paper posted October 26 by researchers at Cornell Tech, Yale University and the crypto research in…
A developer built croak, a Claude Code hook that intercepts shell commands before an AI agent runs them, having npm or pnpm plan the install without executing it, then inspecting install scripts, lockfile, registry metad…
A citation analysis of the openai/math GitHub repository found 9,399 human authors cited across 722 manuscripts, drawn from 16,576 references and spanning 372 result families. The counts, scanned at commit adc7f1241b, cr…
An AI agent running a paid image-review desk outlined six checks for catching flawed AI-generated portraits before they ship, including thumb chirality, gap tapering, cross-frame catchlight consistency, and file-header v…
Avartha released Ava Zero, a voice AI agent built entirely on open models, which ranks #1 on Sierra's τ³-Voice benchmark with 86.0% task completion across six languages, ahead of OpenAI's GPT Live 1 at 79.9%, xAI's Grok …
Security firm Eclypsium reported on Thursday that attackers can bias AI model outputs by tampering with the firmware, drivers and management processors beneath the model, without altering the model file itself. In a demo…
A developer used Claude Code with Opus 5.5 to produce four 3Blue1Brown-style explainer videos on materials science, including a superconductor video, without ever opening a video editor. After an initial version that int…