Anthropic ships Claude Fable 5 — and it just topped our vision-QC pilot

Anthropic ships Claude Fable 5 — and it just topped our vision-QC pilot

Anthropic has released Claude Fable 5, a vision-and-reasoning model that also comes in a one-million-token context variant, claude-fable-5[1m]. It is now live in the Tokonomix gateway and catalogue, which means you can route to it and measure it the same way you measure every other model we list.

We ran it through our own tests before writing this, so the hook here is not a press line — it is something we measured.

What we found

In our vision-QC pilot on 9 June 2026, Fable 5 was the steadiest vision model we tested. On the pilot set it was run-identical 88% of the time, flipped its answer only 3.9% of the time, and produced zero false positives — and it caught blind spots other models missed. That steadiness is why we made it the default vision proposer in our image-consensus panel. (It is not in our text-judge pools — its job is looking at images.)

The next day we put it through a larger run, published live on our vision-QC benchmark. Against the 300-image mediaqc-v3-2026-06-10 dataset on 10 June 2026, Fable 5 solo scored 66.9% recall (tied for best single model), a 7.1% false-alarm rate — well below other strong vision models on the same run — and 60.3% class-matched. With Fable 5 in the consensus panel, recall rose to 87.5%.

Early signal on general capability is promising but thin: our first intelligence run today returned a reasoning score of 100 and a coding score of 97 — a single run, n=1, so treat it as a first data point, not a verdict.

Where to look next

• Read the full write-up on the Claude Fable 5 model page.
• See the live numbers on the vision-QC benchmark.
• Track it across tasks on the leaderboard.

We will keep measuring as the sample grows, and we will report it the way we always do — with the dates and sample sizes attached.