220 artigos por Marcelo Santiago
Artificial Analysis launched VIALS-AA, an implementation of Handshake AI's VIALS benchmark, which tests whether models can interpret scientific images and answer based on the evidence.
Thinking Machines Lab announced Inkling 2 Small with 594 billion parameters and 17 billion active. The model scored 71.8% on EnterpriseBench-CoreCraft and 46.8% on EnterpriseOps-Gym.
OpenAI launches BioBasicBench, a task set to measure AI agents in biology scenarios. What the benchmark evaluates and how it compares to existing sets.
Google adds voice mode to Antigravity, letting developers guide code agents by audio. What does this change in the development workflow.
600,000 Nvidia H200 chips are awaiting release in China. U.S. export controls keep the stockpile in Hong Kong.
Seven checks for comparing agent benchmarks: task unit, sample size, generalization, test conditions and score limits.
NVIDIA adds four Cosmos3-Nano checkpoints for automotive and robotics simulation. See capabilities, requirements, license and limits.
AI2 introduces ToyBlocks, a simulation environment with four tasks to study generalization and physical transfer in robotics.
Codex 0.167.0 adds an agent hub in the terminal, highlights unread results and preserves drafts when switching backends. See what changes.
On the tau2-Bench Airline leaderboard from November 6, GPT-5.1 scored 80.0%. Learn about the method, minimum sample size and limitations before choosing an agent.
Anthropic analyzed 207,049 scientific sessions in Claude and Claude Code and separated observed usage from productivity projections.
Learn about NVIDIA Auto Ontology, an Apache 2.0 repository for giving semantic context to agents and the security considerations before adopting it.
In-depth technical curation: agentic architectures, real token benchmarks, MCP servers and open-source tools in your inbox and on the blog.
