According to VisionWatch Beating monitoring, the third-party assessment agency Artificial Analysis has released the Claude Opus 5 score. It scored 61 in the overall intelligence index based on 9 tests, narrowly ahead of Fable 5 with 60 points. GPT-5.6 Sol scored 59 points, and Kimi K3 scored 57 points.
Opus 5 had an average cost per task of $2.03, 26% lower than Fable 5's $2.75. It topped the rankings in the GDPval-AA v2 and AA-Briefcase knowledge work assessments, and jointly secured the top spot in the Programming Agent index when paired with Claude Code. The Terminal-Bench v2.1 scored 89%, roughly equaling GPT-5.6 Sol.
The model provides 5 levels of reasoning intensity. From low to max, the Token output differs by about 8 times, and the GDPval-AA v2 score differs by 407 Elo. Users can exchange more Tokens for stronger performance or actively lower costs.
However, weaknesses are also apparent. Opus 5's factual knowledge still lags behind Fable 5. In the AA-Omniscience test, its illusion rate has increased to 50%, 14 percentage points higher than Opus 4.8. The cost-effectiveness of the low reasoning levels is also slightly lower than the GPT-5.6 series.
