Re:November 2026 Claude Fable 5.2 Scores 65 on Artificial An...
by nl on Tuesday November 10, 2026 @06:06PM
(#50447601)
> When comparing at more reasonable medium or high levels of reasoning, the gap between Opus and Sol narrows considerably. Opus being about 2 higher than Sol at the same reasoning level. However, Sol 6.1 xhigh matches Opus 5.5 high on scores (52), but while still being slightly faster and costing 78% less ($0.76 versus $3.46).
> https://artificialanalysis.ai/models?models=claude-opus-5-5-...
Yes, but my point is that these are non-coding benchmarks at these non-max levels.
On coding at non-max Opus really does seem noticeably better, and my subscription seems to go longer. Look at the other comments in this thread - every single one says the same thing.
I think it is a narrow view to only consider the AA coding agent index as the only metric, especially with the models running at max.
For the link you shared, Opus used 3.7x as many tokens, and 2.5x the time to get its score. In the real-world building things with these tools, that kind of increase in time and cost (or subscription usage) matters, at least to me.
When comparing at more reasonable medium or high levels of reasoning, the gap between Opus and Sol narrows considerably. Opus being about 2 higher than Sol at the same reasoning level. However, Sol 6.1 xhigh matches Opus 5.5 high on scores (52), but while still being slightly faster and costing 78% less ($0.76 versus $3.46).
https://artificialanalysis.ai/models?models=claude-opus-5-5-...