Unverified50% confidenceFactExact time
A 360M-parameter code model trained by the trifero.ai lab under Strife Technologies initially reported 26.8% pass@1 on HumanEval, and after removing instruction data overlapping with the benchmark, the true number was 22.6%.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/13/2026
First Seen
Valid until: 1/11/2027
Sources
Related Entities
Related Claims
UnverifiedCRN v2 has approximately 34 million trainable parameters, equivalent to 0.73% of the 4.65 billion-parameter text module.68% similarUnverifiedIn the experiment, the 27B model scored 61.0% on LiveMath when skills were written by the 4B model, whereas it scored 56.3% when it wrote the skills itself67% similarUnverified初版 Codex 采用约120亿参数规模，在 HumanEval 基准测试中达到 28.8% 的 pass@1 通过率67% similarUnverifiedLoRA使可训练参数量通常仅为全参微调的0.1%-1%，可在单张RTX 3090/4090（显存24GB）上微调7B乃至13B级别的模型67% 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.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1013053API
curl https://kongchang.com/api/v1/knowledge/claims/1013053MCP
get_claim(id=1013053)