Unverified50% confidenceBenchmarkExact time
在 GSM8K 基准上，标准思维链准确率约为 56%-63%（GPT-3 规模），PAL 可提升至 72% 以上
1
Sources
50%
Confidence
Long-term
Relevance
7/6/2026
First Seen
Sources
Fixing the Three Big Weaknesses of LLMs: Knowledge, Computation, and Tool Use
bilibili晴天AI实战7/4/2026
Related Claims
VerifiedChain-of-Thought prompting improved PaLM 540B's accuracy on the GSM8K math benchmark from 17.9% to 58.1%.72% similarUnverifiedThe MTP-LX 4-bit version scored 55 out of 100 on a financial planning reasoning task as rated by GPT-5.5 Thinking, while GPT-5.5 Pro scored 82 on the same question.70% similarUnverified在GPT-4发布初期，主流模型在GPQA上的准确率约为35-40%，人类领域专家基准线约为65%，随机猜测为25%70% similarUnverifiedGSM8K发布时GPT-3的zero-shot准确率仅约5%，加入few-shot示例也不超过35%70% similarUnverifiedOn the coding index (reasoning mode) in third-party evaluations by Artificial Analysis, GPT-5 Mini High scored 9 points higher than Haiku 4.569% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/112765API
curl https://kongchang.com/api/v1/knowledge/claims/112765MCP
get_claim(id=112765)