Unverified50% confidenceBenchmarkExact time
In the third round, with the model running alone, the Shell evaluation set score rose from 42/71 for the base model to 61/71, the Korean set went from 27/30 to 27/30 (unchanged), and the end-to-end Shell evaluation set went from 69 to 70.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/3/2026
First Seen
Valid until: 2/1/2027
Sources
Related Entities
Related Claims
UnverifiedOn the average MUSIQ score of nine-frame sampling, the frame-extraction input was higher than the screenshot input on all three models: Kling 4.0 at 62.60 vs 58.92, Wan 3.0 Prime at 67.97 vs 53.70, and Seedance 2.5 at 59.40 vs 46.35.65% similarUnverifiedKimi K3 scored 77.8% on coding benchmarks, ahead of OpenAI's new model at approximately 77.6% and another competing product at approximately 76.8%.63% similarUnverifiedIn the experiment, Level 3 (full checklist prompt) scored an average of 30.3/40 (76%), higher than Level 1's 24.7/40 (62%) and Level 2's 22.2/40 (55%).63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1130053API
curl https://kongchang.com/api/v1/knowledge/claims/1130053MCP
get_claim(id=1130053)