Test shows Opus 5.5 writes the same program in half the lines of Astra
scaling01 · x · 2026-09-28
Peter Gostev proposes an alternative benchmark: who can accomplish the same task with the fewest lines of code? In his tests, Claude Opus 5.5 uses roughly half the lines of code that Astra does, with no noticeable drop in physics or visual quality. The goal is to approximate "elegance" in AI-generated code — fewer lines often means a more generalized approach — though he admits the proxy is imperfect.
More from coding & agent
- Stanford puts all 9 lectures of CS329A: Self-Improving AI Agents online for free — ghumare64 · 2026-09-28
- Agents are getting good at optimizing inference server configs, dev shows — charles_irl · 2026-09-28
- François Chollet: Hand-writing code no longer pays off, LRMs unlock better workflows — fchollet · 2026-09-28
- Anthropic open-sources its full prompt engineering course, now 38K stars on GitHub — victor_explore · 2026-09-28
- Salesforce's Critical-State RL Gains ~14 Points on BFCL v4 by Training Only the Pivotal Tool Call — dair_ai · 2026-09-28
- AI code-sandbox provider Daytona officially moves to new domain — glcst · 2026-09-28