Unverified50% confidenceBenchmarkExact time
On FastBench, Qwen3-VL-8B improves from 32.9% at 2 FPS to 44.6% at 24 FPS, but as history is compressed, the gains gradually saturate.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/13/2026
First Seen
Valid until: 10/27/2026
Sources
Related Entities
Related Claims
VerifiedQwen3 27B scored 90.3 on LiveCodeBench V6, slightly surpassing Opus 4.6 Max's score of 88.8.72% similarUnverifiedQwen3 27B's score on the long-horizon autonomous coding metric DeepSWE jumped from 13.3 in the previous generation to 42.2.70% similarUnverifiedOn a single RTX PRO 6000, peak VRAM for Qwen3.8-27B-NVFP4 dropped from 72.9 GB to 40.2 GB, and per-step time was reduced by approximately 11%69% similarUnverifiedWhen deploying Qwen3 27B with 16 GB of VRAM, if some weights are offloaded to system memory, the speed drops sharply from tens of T/s to single digits.69% similarUnverifiedGPT-OSS-20B BF16 loading was reduced from 545 seconds to 70 seconds (7.8× speedup), and Qwen3.5-397B achieved a 1.93× to 2.3× improvement.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1013401API
curl https://kongchang.com/api/v1/knowledge/claims/1013401MCP
get_claim(id=1013401)