Tip: Opus 5.5 at max thinking burned its whole 128K output budget on a simple SVG task
Simon Willison·medium signal
Simon Willison moved his Claude Code and Codex defaults to Opus 5.5 and GPT-6 Sol. He reports that Opus 5.5 at its maximum thinking level exhausted the 128,000-token output limit while reasoning through a simple SVG drawing. Anthropic lists effort levels low through max with medium as the default. Keep agent loops at medium or high and treat max as a per-task escalation. He also moved his Datasette Agent demo to GPT-6 Luna ($0.10/$0.50) and found it competent at SQL and HTML/JS.