Unverified50% confidenceBenchmarkExact time
On the turn-end detection task, Marimba reaches an AUROC of 0.935, while the comparison models (Qwen3-Omni-30B-A3B, Qwen2-Audio-7B, FireRedAudio 10.6B) range between 0.54 and 0.79.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
11/17/2026
First Seen
Valid until: 12/1/2026
Sources
Related Entities
Related Claims
UnverifiedMarimba's turn-end detection readout reaches AUROC 0.930, while the dedicated model Smart Turn v3.2 achieves 0.982 on its own test set.74% similarUnverifiedQwen3.5-122B-A10B produced 8 refusals in 100 tests, with a KL divergence of 0.0856.67% similarUnverifiedTev1 and the locally run Qwen3.5 9B showed no statistically significant difference in mean field accuracy (0.917 vs 0.918, p = 0.26).67% similarUnverifiedOn "continuation of previous operation" type utterances, Qwen3.6-35B-A3B achieved an accuracy of 86.3%, while the 4B and 8B models achieved only 70.4%.66% similarUnverifiedAfter applying WavePrune to Qwen3-8B, the HELMET score improved from 35.7 to 40.0.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1179997API
curl https://kongchang.com/api/v1/knowledge/claims/1179997MCP
get_claim(id=1179997)