Box says Opus 5.5 cuts token usage 63% and runs 30% faster than Opus 5
bcherny · x · 2026-09-24
Box has been testing Claude Opus 5.5 on complex enterprise knowledge work with unstructured data via Box Agent, reporting frontier-level capability with major gains over Opus 5: 63% fewer tokens, 42% less verbosity, 30% faster, plus a cheaper per-token price (Anthropic says it costs 40% less to run than Opus 5).
Task highlights:
- Financial services due diligence (+39% accuracy): finding every miscalculation in an acquisition target's pricing tool from a year of transaction records — perfect score on every attempt, with 82% fewer tokens.
- Anthropic's own announcement: Opus 5.5 is the first model in the Claude 5.5 family, performing at Claude Fable 5.1 level for most tasks.
Box frames it as a big win for agentic computer use, coding, analytics and data workloads.
More from Companies & People
- Alpha School's outdoors campus opens: 2 hours of AI academics a day — seanmcdonaldxyz · 2026-11-11
- Meta AI web visits hit record 43.93M in October, up 27.67% MoM and 94.74% YoY — gaganghotra_ · 2026-11-11
- AI safety researcher hires paid, credited RA/co-writer to speed up memos and blog posts — davidmanheim · 2026-11-11
- Model PM jobs analyzed: 76% want evals skills, only 2% ask for a PhD, median base $345K — aakashgupta · 2026-11-11
- Everyone fears AI labs eating SaaS—Theo is moving faster than the labs in every direction — weswinder · 2026-11-11
- Baidu World 2026 kicks off tomorrow with 'Intelligent Organizations' theme — Baidu_Inc · 2026-11-11