Unverified50% confidenceFactExact time
The performance degradation caused by lost formatting cues is reproducible across six answer models, covering five models across the Claude Opus, Sonnet, and Haiku tiers plus OpenAI's GPT-5.6.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/22/2026
First Seen
Valid until: 1/20/2027
Sources
Related Entities
Related Claims
UnverifiedIn the tests, each Claude model (Haiku 4.5, Haiku 5.5, Sonnet 5.5, Opus 5.5) exhibited outcome-bias-induced score leakage on at least one case.77% similarUnverifiedThe two-stage classifier runs on top of Claude Sonnet 5.5, Opus 5 and 5.5, and Fable 5 and 5.1, and Anthropic says its false positive rate is below 0.1%.69% similarUnverifiedIn the honesty test, the number of confident wrong answers was: Opus one, Sonnet four, DeepSeek Flash eighteen, and DeepSeek Pro twenty-six.67% similarUnverifiedThe blogger judges that Haiku 5.5's absolute output quality does not match flagships such as Sonnet, Opus, or GPT-6 Astra, but delivers capability close to the frontier level of half a year ago at one-quarter of the price.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1084059API
curl https://kongchang.com/api/v1/knowledge/claims/1084059MCP
get_claim(id=1084059)