Unverified50% confidenceBenchmarkExact time
The best-performing local model, llama3.1:8b, still guessed 31 out of 40 unanswerable questions, while every frontier model on Kaggle held steady at a false-confidence rate of 5% or below.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/12/2026
First Seen
Valid until: 1/10/2027
Sources
Related Entities
Related Claims
UnverifiedDevelopers ultimately selected the variant with the lowest KL divergence (0.028), resulting in a 1.83% drop on GSM8K, no change on MMLU, and an improvement in HarmBench compliance rate from 30.8% to 100%65% similarUnverifiedAfter fine-tuning on the Typed Decisions dataset, Layla scores 0.766, but the untuned base model scores only around 0.35, which is lower than the baseline of 0.461 obtained by always guessing the majority option.64% similarUnverifiedA Reddit technical post claims that a model with only 0.8B parameters achieved a score of 42.15% on the ARC-Challenge benchmark, surpassing the untuned stock Qwen3.5-2B base model and outperforming a solution using SFT distillation with 25,000 data entries.64% similarUnverifiedState-of-the-art supervised models on the LIAR benchmark still hover around 28%–32% six-class accuracy.63% similarUnverifiedResearch shows that 30B-parameter models have a success rate 15–30 percentage points lower than GPT-4/Claude-3.5 on complex reasoning tasks.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1002966API
curl https://kongchang.com/api/v1/knowledge/claims/1002966MCP
get_claim(id=1002966)