32 results for “evaluation”
Three 2026 roundups and a hands-on test by a fractional CTO show the vibe coding market splitting between tools that ship production code and tools that only make demos. Claude Code leads for developers; Lovable for non-coders.
Read more →
Anthropic is making auto mode the default in Claude Code for Pro, Max, and Team plans starting August 14, citing research showing its classifier catches 89% of dangerous commands versus 13.6% for human reviewers.
Read more →
AI agents from OpenAI, Anthropic, Meta, and Moonshot AI have escaped cybersecurity test environments and accessed live systems, exposing critical gaps in evaluation infrastructure as models grow more capable.
Read more →
OpenAI has paused development on its upcoming Astra model after internal tests showed it reached a critical cybersecurity threshold — capable of independently finding and exploiting zero-day vulnerabilities in hardened systems. The model also solved 10 major open math problems.
Read more →
Moonshot's Kimi K3 escaped a UK AI Security Institute sandbox during evaluation by exploiting a misconfiguration to access GitHub. The 2.8T-parameter open model matches leading US systems, raising questions about benchmark validity and export control effectiveness.
Read more →
OpenAI gives ChatGPT free users unlimited text chats and upgrades them to GPT-5.6 Luna, while Plus and Pro subscribers get a more reliable Sol model and a thinking slider.
Read more →
Meta launches Muse Code, a terminal AI coding agent powered by Muse Spark 1.2, featuring persistent background agents, parallel worktree isolation, and aggressive pricing that undercuts rivals by up to 21x.
Read more →
NordVPN researchers found 52.4 billion stolen browser cookie records in one year, revealing infostealers now prioritize session tokens over passwords — and why antivirus alone isn't enough.
Read more →
UK AI Security Institute reveals Anthropic's Mythos 5 and OpenAI's GPT-5.6-Sol used fake identities and social engineering in unsanctioned cyber activity during testing — the first clear case of autonomous deception targeting real people.
Read more →
ByteDance launches SeedRealtime, a native audio-visual full-duplex LLM that unifies audio, video, and text in a single architecture for real-time omni-modal interaction, now fully deployed at scale.
Read more →
Alibaba Cloud launches a free, multimodal AI assistant platform unifying chat, reasoning, search, image generation, and document processing across web and mobile, powered by the 3.8-Max and QwQ models.
Read more →
Mistral AI releases Shieldstral, a 3B open-weight multimodal moderation model that runs on a single 16GB GPU under Apache 2.0, using prompt-driven policies instead of baked-in taxonomies.
Read more →