Artificial Analysis ranks Claude Fable 5.2 first at 65, with 3.8x median tokens

Artificial Analysis, an independent benchmarking site, lists Claude Fable 5.2 at 65 on its Intelligence Index, first of 74 models, according to the model page it published on 9 November 2026. The page gives a median of 40 for the model class and names no other model's score. It is the first independent result I have seen for the model Anthropic released the same day, whose own benchmark figures are covered in our launch report.

What the page reports

• Configuration tested: Fable 5.2 at Max effort.
• Intelligence Index: 65, rank 1 of 74.
• Price: $8.00 per million input tokens and $40.00 per million output tokens.
• Speed: 101 output tokens per second.
• Context window: 1M tokens.
• Output tokens used to run the index: 275M, against a class median of 72.8M.
• Cost: the page gives $26.54 per Intelligence Index task and does not state a cost for the full run. I could not confirm from the page exactly how that per-task figure is defined.

Why the token count matters

Output tokens are billed. A model that uses 275M tokens where the median model uses 72.8M, about 3.8 times as many, can cost more per finished task even when its per-token price is lower than its predecessor's. Anthropic's launch post reports a price of $8 per million input tokens, which TechCrunch reports is down from $10.

A commenter on the Hacker News thread (item 50431144; the site is not on this publication's source list, so it is not linked), which had 102 points and 53 comments when I read it, made that point sharply. The commenter wrote that Fable 5.1 was already the most expensive model on the site's cost-per-task chart and that 5.2 doubles it, and said output tokens per task rose from 118k (46k answer, 71k reasoning) to 274k (81k answer, 192k reasoning). That is one commenter's reading of the site's charts; I could not check the 5.1 figures against a page I fetched.

Another commenter argued the opposite for latency: that Fable 5.2 at low and medium effort sits on the site's frontier of time per task against intelligence. Both can be true, because effort setting changes the token count. The result I could verify is for Max effort only.

What it does not tell you

• The index is an aggregate. It does not show how the model does on your documents, tools or ticket types.
• The page names no comparison models in the text I read, so how far ahead 65 is of the next model is unknown to me.
• Artificial Analysis can revise scores as it adds tests, and I have only the figures on the page at the time I read it.

What to do

If you are evaluating Fable 5.2 for an agent or coding workload, run a sample of your real tasks at two effort levels and record output tokens, dollars and pass rate per task. Compare that to your current model on the same sample. Set a token cap per task before rollout, since a model that reasons longer can quietly raise a monthly bill that was forecast from per-token prices.