x.com/samsinai/status/2102485970239320560
1 correction found
it crushed the performance ceiling beyond all humans.
The benchmark shown does not compare Opus 5.5 to all humans. Anthropic’s underlying RNA design benchmark is against 57 human participants, with the top comparison point defined as the best participant in that sample.
Full reasoning
Anthropic’s own materials do not show that Opus 5.5 performed beyond all humans.
What Anthropic says about this evaluation:
- For Opus 5.5, Anthropic says the biology result came from a "long-horizon molecular prediction and design evaluation conducted in collaboration with Dyno Therapeutics."
- In Anthropic’s documentation for the same Dyno RNA design benchmark, the task is described as being benchmarked against 57 human participants drawn from the leading edge of the U.S. ML-bio labor market.
- Anthropic also defines the upper human benchmark as the model exceeding "the top human participant" in that evaluated cohort.
So this result can support a claim like "Opus 5.5 beat the best human in this benchmark sample". It does not support the stronger universal claim that it went beyond all humans.
That distinction matters because the benchmark is a limited comparison to a specific set of 57 participants, not humanity as a whole.
2 sources
- Claude Opus 5.5 | Anthropic
Anthropic says: “Opus 5.5 achieved improvements on a long-horizon molecular prediction and design evaluation conducted in collaboration with Dyno Therapeutics.”
- System Card: Claude Opus 5 - systemcards.org
The extracted Anthropic system card says the black-box RNA sequence design challenge was “benchmarked against 57 human participants” and defines the second benchmark as exceeding “the top human participant.”