Unverified50% confidenceBenchmarkExact time
在 Terminal Bench 2.1 测试中，AutoPrompt Scale 使 OpenCode 在 89 个任务中的失败数从 29 降到 16，多解决 13 题，分数提升 14.61 个百分点
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/27/2026
First Seen
Valid until: 11/25/2026
Sources
Related Entities
Related Claims
UnverifiedOn the SWE-Bench programming benchmark, Claude Code scored 80.8%, while GitHub Copilot scored only 72.5%.62% similarUnverifiedMiniMax M3 received an average score of 58.3 across 7 high-difficulty coding tasks in a systematic evaluation61% similarUnverifiedIn practical tests, completing a search task using AnySearch Skill consumed 18 quota points, while not using it consumed 23 quota points, saving approximately 27%.60% similarUnverifiedThe median prompt length for SWE-Bench Pro is 30,098 characters, while the median for the task description portion of Frontier Code is only 9,082 characters.59% similarUnverifiedIn the Terminal Bench 2.1 test, GPT-5.6 Ultra scored close to 92%, Fable 5 scored 88%, which are essentially on par considering the margin of error.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/807691API
curl https://kongchang.com/api/v1/knowledge/claims/807691MCP
get_claim(id=807691)