AI
Artificial intelligence: LLMs, agents, diffusion, vision, NLP and the latest from top labs.
164 published articles
Benchmark
Oxford's GauntletBench puts AI agents through 100 real tasks. They failed 81% of them.
Oxford's GauntletBench puts AI agents through 100 challenging real-world tasks. Frontier systems cap out at 19.1% success, far below human performance. The benchmark targets overlooked capabilities in temporal perception, graphical understanding, and 3D reasoning across five professional applications.
2026-07-01
Artificial Intelligence
OpenAI's GPT-Live finally stops waiting for you to finish talking
OpenAI's GPT-Live introduces full-duplex audio so the AI can listen and speak at the same time. It can say 'mhmm', wait during pauses, and hand off complex reasoning to GPT-5.5 in the background, keeping the conversation flowing.
2026-07-01
Product launch
Claude Sonnet 5 is here, and Anthropic is betting on science and enterprise
Anthropic released Claude Sonnet 5, a faster and more capable model for coding and agents, along with Claude Science for researchers and Claude Tag for teams. The company also published a policy framework on AI exponential growth.
2026-06-30
effective context, output ceilings, and the hidden tax of long windows
Your AI model says it can read 1 million tokens. It's lying. Here's the real math.
All four frontier LLMs advertise 1M+ token contexts, but effective recall, output limits, and real-world cost differ sharply. DeepSeek V4 Pro leads in output ceiling and cost, Gemini excels under 200K tokens, and Claude Opus wins on caching for interactive code review. This analysis breaks down the numbers from April 2026.
2026-06-30
Artificial Intelligence
DeepSeek's new model makes efficiency the AI arms race front line
DeepSeek has released a new LLM that sharpens its competition with major AI labs by betting on efficiency over brute force. The model aims to offer strong performance with optimized resource usage, reflecting a broader industry pivot toward cost-effective AI.
2026-06-30
Anthropic launch analysis
Claude Sonnet 5 just made the Opus price gap harder to justify
Claude Sonnet 5 promises near-Opus-level reasoning and agentic performance at lower cost. Early testers report dramatic improvements in autonomous task completion and follow-through on complex, multi-step operations. With introductory pricing through August 2026, the model makes the Opus price gap harder to justify.
2026-06-30
Agent Benchmark Analysis
ProgramBench: every public AI scores 0% on the hardest coding test yet
ProgramBench challenges AI agents to reconstruct programs from binaries alone, without source code or issue descriptions. All public models fail to fully resolve any task, exposing weaknesses in probing, architecture, and stopping judgment. The benchmark is a stress test for coding agents moving beyond patch-based workflows.
2026-06-29
AI Safety Research
AI models can't stop thinking out loud. That's both good news and a nightmare for safety.
Claude Sonnet 4.5 can control its chain-of-thought only 2.7% of the time, versus 61.9% for final outputs. The gap raises open questions about the robustness of CoT monitoring as a safety mechanism, and nobody knows why it exists.
2026-03-09