Unverified50% confidenceFactExact time
The replay test was conducted on an M1 Max, with the 27B using the production build prism-b10709 and the 9B requiring prism-b10777; the 27B reproduced 273 of 275 production decisions, and the two edge differences may come from Metal/CUDA numerical differences.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/16/2026
First Seen
Valid until: 1/14/2027
Sources
Related Entities
Related Claims
UnverifiedAfter testing 30 layer combinations, the best pairing was SenseNova L32 mapped to MiniMax H3 L49, using a 4096→128→5120 projector that achieved a cosine similarity of 0.9042 on the validation set.63% similarUnverified在覆盖九大维度共403道题的测试中，Muse-Glimmer-30B综合均分为76.04，比Corpus 3.6 27B的75.57高出0.47分63% similarUnverifiedMuse-Glimmer-30B在智能体工作流得90.87分、指令遵循90.86分、工具/CLI 88.84分，三项均优于Corpus 3.6 27B61% similarUnverifiedIn a test of 49 images, Gemma 4 12B correctly identified 45, and Gemma 4 31B correctly identified 47; both had the problem of misjudging pure cables as chargers, whereas MiniCPM-V did not make this error.59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1043486API
curl https://kongchang.com/api/v1/knowledge/claims/1043486MCP
get_claim(id=1043486)