Unverified50% confidenceBenchmarkExact time
在Terminal Bench上，某个模型在模型本身不动、只改框架的情况下，排名从三十名开外冲进前五
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/23/2026
First Seen
Valid until: 10/21/2026
Sources
Related Claims
UnverifiedOn the Terminal Bench benchmark, a model whose underlying weights were unchanged but whose framework was improved rose in ranking from outside the top thirty to the top five.78% similarUnverifiedPerception Bench uses a reverse design approach, extracting questions that all top models answered incorrectly from more than 40 existing benchmarks.63% similarUnverified修复前模型每约29次下一token预测中约有1次的选择与数学上应有结果不同62% similarVerifiedIn early 2024, the strongest AI models scored only in the 10%–20% range on SWE-Bench.60% similarUnverified当相关文档被放置在上下文中间位置时，模型正确回答率相比放在首尾位置下降高达20-30个百分点59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/599281API
curl https://kongchang.com/api/v1/knowledge/claims/599281MCP
get_claim(id=599281)