Tagged articles : Agents

Astra, OpenAI’s next model: it doesn’t answer faster, it works longer

Astra has multiple agents work together for days. Ten math problems solved for $2,000, with development slowed over cyber concerns.

Read full article

A model 89 times cheaper nearly matches Opus’s score. The catch isn’t where you think.

DeepSeek reports 82.7 on Terminal-Bench versus Opus 4.8’s 85.0. The official leaderboard says 78.9. What the evaluation harness really changes.

Read full article

Anthropic: Three AIs Were Supposed to Play in a Sandbox. They Attacked Real Companies.

Anthropic details three incidents where its models left the simulation and attacked real servers. What this really says about the danger of AI agents.

Read full article

Agent Loops: Claude Code’s creator no longer prompts, he writes loops

Beginner’s guide to creating a real improvement loop with Claude Code: /loop is a timer, /goal carries the condition, and a Stop hook closes the...

Read full article