Studio blog

Notes from the workshop

Write-ups on benchmarks, models, and agent harnesses, and what we’re learning as we build.

How Good Are LLM Agents At Implementation?

Who wins when the task is clear, and efficiency is the name of the game?

Fable 5 ties Opus 4.8 on structured implementation work — at 1.6x the price. Sonnet 4.6 sits a full tier below.

The benchmark's first flight — and why a fancier subagent setup didn't beat a single agent.