Opus 5.5 uses 26k tokens vs Astra's 12k yet costs 23% less per task at equal AA score
ChrisGPT · x · 2026-10-06
ChrisGPT asks whether token efficiency matters more than cost per task at equal intelligence. His data point: Opus 5.5 medium uses 26k output tokens vs Astra high's 12k, but both score 51 on AA—yet Opus costs $1.34/task vs Astra's $1.73, 23% cheaper. His take: at the same intelligence level, you'd rather spend more tokens and pay less per task; per-token efficiency is the wrong optimization target for cost-conscious teams.
More from Models
- Opus 5.5 is efficient on subscription, not via API — 6.1 remains the workhorse — haider1 · 2026-10-06
- Hiding Y-Axis Labels in Early nanogpt Benchmarks Is "Academic Dishonesty" — PMinervini · 2026-10-06
- LLM MoEs run at ~5% sparsity, cited as counterexample in consciousness complexity debate — JoshPurtell · 2026-10-06
- Report: Zhipu's GLM 5.3 Also Hit a Delayed Release — teortaxesTex · 2026-10-06
- Is There a Market for the 10th-Best Open Model? $5B Capex Question Sparks Debate — ericjang11 · 2026-10-06
- How AA Benchmarks 26 Search API Products Across 13 Providers — ArtificialAnlys · 2026-10-06