Unverified50% confidenceFactExact time
The test is based on the Qwen3.8-27B model, using EXL3 4.0bpw quantization, a DFlash2 block drafter (EXL3 4bpw, 7-token draft), greedy decoding, running on 2 T4 GPUs.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/25/2026
First Seen
Valid until: 1/23/2027
Sources
Related Entities
Related Claims
UnverifiedThe test was conducted based on the Qwen3 27B NVFP4 quantized version released by Unsloth.77% similarUnverifiedThe study tested four sizes of the Qwen3 series—0.6B, 1.7B, 4B, and 8B—running at FP16 precision, greedy decoding, a single frozen prompt, and no fine-tuning.75% similarUnverifiedIn actual tests of 8-bit and 4-bit (OQ4E) quantization of the same Qwen3 27B model based on the MLX framework, each won 5 out of 10 accuracy tests, resulting in a tie.71% similarUnverifiedBilibili developer Luke conducted a cross-evaluation test of the Qwen3 series 27B model released by Unsloth across multiple quantization levels from Q1 to Q8.71% similarUnverifiedThe locally deployed Qwen3.6-35B-A3B model once caught real bugs in code originally written by Claude Sonnet/Opus, and the author emphasized this was not a benchmark test but happened by chance.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1096596API
curl https://kongchang.com/api/v1/knowledge/claims/1096596MCP
get_claim(id=1096596)