AI BriefingHaiku overtakes Opus
Key point
Across 880 evaluations, Haiku 4.5+skill outperformed the Opus 4.7 baseline.
Details
Through 880 evals, 11 skills × 8 models × 5 scenarios were compared, each skill tested both with and without the skill applied.
The key result is that Haiku 4.5 baseline 61.2% rose sharply to Haiku 4.5 + skill 84.3%. Under the same conditions, Opus 4.7 baseline was 80.5%.
Cost efficiency also stood out.
- Haiku + skill: $0.12 per run
- Opus baseline: $0.61 per run
- The added cost of putting the skill into Haiku is about 1.5 cents
- Putting the same skill into Opus added 39 cents per run
The effect of skills was confirmed across all vendors, and weaker models showed larger gains. Haiku recorded +23.1p, while Opus 4.7 recorded +14p.
In practical terms, this suggests that for repetitive tasks like commit messages, code review, and refactor suggestions, the combination of Haiku + a good skill is fast and sufficiently accurate. Codex variants and Cursor Composer-2 also showed improved performance when skills were applied.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
US Appeals Court Upholds Pentagon Blacklisting of Anthropic as Supply Chain Risk
Anthropic Pacing to $100B ARR; Rumors of OpenAI Solving Navier-Stokes Spark Controversy