Unverified50% confidenceBenchmarkExact time
On performance model code construction tasks, the average pass rate of all model configurations other than GPT-5.6 Sol was below 15%, with extremely large fluctuations across multiple runs.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/11/2026
First Seen
Valid until: 12/10/2026
Sources
Related Entities
Related Claims
Unverified权重绑定在中小规模模型上(如GPT-2 124M)通常能提升性能并减少约30%的嵌入相关参数，但在超大规模模型中收益递减74% similarUnverified研究表明，即使在 GSM8K 或 MATH 数据集上接近满分的模型，面对稍作变形或需要跨领域组合的问题时性能下降可达30-50个百分点72% similarUnverifiedFor structured tasks such as code completion and function rewriting, the output quality gap between mid-to-low-end models and top-tier models is typically within 5%.72% similarUnverifiedExperiments conducted by the research team on models such as GPT-2 showed that after only 5–9 generations of recursive training, the quality of model outputs degrades significantly.70% similarUnverifiedEven GPT-4-level Agents often achieve end-to-end success rates below 60% on complex tasks involving more than 5 tool calls.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/902747API
curl https://kongchang.com/api/v1/knowledge/claims/902747MCP
get_claim(id=902747)