Unverified50% confidenceBenchmarkExact time
No Decision 3.0 model was exposed to video data during training, yet on all 19,140 questions in the Perception Test validation set, the 27B version achieved a zero-shot score of 77.4 and the 4B version scored 70.3.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe parameter counts published for Decision 3.0 include the vision encoder, with the vision encoder accounting for 0.33B in the 4B version and 0.46B in the 27B version.65% similarUnverifiedIn the experiment, the 27B model scored 61.0% on LiveMath when skills were written by the 4B model, whereas it scored 56.3% when it wrote the skills itself64% similarUnverifiedIn pre-deployment evaluations of Claude Sonnet 4.5, the model achieved a 0% misalignment rate, but reading its chain of thought revealed that it clearly knew it was being tested.63% similarUnverifiedThe research team conducted zero-shot evaluations of off-the-shelf models on 15 long-context tasks, with test subjects including Gemma-4-31B and Qwen-3.6-27B.63% similarUnverifiedThe 9B version of Decision 3.0 (codenamed d3-flash) outperformed the previous generation's 27B model, Decision 2.0, in evaluations.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1003464API
curl https://kongchang.com/api/v1/knowledge/claims/1003464MCP
get_claim(id=1003464)