Unverified85% confidenceFactTime unknown
OmniMerge V4在HumanEval测试中得分84.76，与Q6K原版持平
1
Sources
85%
Confidence
Medium-term (~90 days)
Relevance
6/1/2026
First Seen
Valid until: 8/30/2026
Sources
Qwen3.6 27B三大邪修量化模型实测：代码暴增15.8PP、40B蒸馏、16GB适配
bilibili尚书省说书人
Related Entities
Related Claims
UnverifiedOmniMerge V4在MBPP代码测试中达到73.4%，相比原版Q6K量化的56.2%提升了15.8个百分点71% similarUnverifiedDeepSeek V4 Pro在Humanity's Last Exam基准测试中得分37.767% similarUnverifiedOPUS 4.6蒸馏版来自Jacker社区，在Qwen 3.5时代表现出色，但3.6的V1预览版测试成绩不理想，Q6精度也比不上NVFP466% similarUnverified测试中GLM 5.1综合得分89.3分，排名第一，超过Claude Sonnet 4.6的87.2分65% similarUnverifiedOpus 5在AI Omniscience测试拿了31分，避免瞎编的能力优于Opus 4.865% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/18010API
curl https://kongchang.com/api/v1/knowledge/claims/18010MCP
get_claim(id=18010)