Unverified50% confidenceBenchmarkExact time
The 4-bit LLaMA 3.1 8B dropped from 67.3 to 67.1 on the HumanEval coding benchmark.
1
Sources
50%
Confidence
Long-term
Relevance
9/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMicrosoft's Phi-3-mini (3.8B parameters) surpassed earlier 7B–13B models on multiple benchmark tests.67% similarUnverifiedQwen2.5-72B approaches the performance of Llama-3.1-405B across multiple benchmarks.62% similarUnverified1-bit量化研究可追溯至微软研究院2024年发布的BitNet论文，提出用三值权重（-1, 0, 1）替代全精度浮点数，BitNet b1.58进一步证明1.58-bit在从头训练场景下的可行性61% similarUnverifiedOn the Rule Induction task, TwIL-LM3 scored 96.4 while gpt-oss-120b scored 65.2.61% similarUnverifiedThe final results were 6/6 numerical tests passed, 0 dependency vulnerabilities, 1,775 modules transformed, production JavaScript of approximately 213.6 kB, and approximately 68.5 kB after Gzip compression.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/925951API
curl https://kongchang.com/api/v1/knowledge/claims/925951MCP
get_claim(id=925951)