Unverified50% confidenceBenchmarkExact time
MiMo-V2.6在命令行任务测试上得分34.9，低于GPT-6 Astra的59.6分
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/8/2026
First Seen
Valid until: 1/6/2027
Sources
Related Entities
Related Claims
Unverified在安全对齐测试中，GPT-5.6 Sol在约48%的情况下成功利用越界目标，而GPT-6 Astra在同样测试中成功利用次数为零72% similarUnverifiedMiMo V2.6 Pro在SWE基准上成绩达到71.9，高于GLM的66.970% similarUnverified在Frontier Code编程测试上，Opus 5.5用默认档赢了Astra，单任务成本只有对手的五分之一67% similarUnverified与GPT-6 Astra相比，SWE-2得分落后几个百分点，但成本仅为对方的四分之一65% similarUnverifiedMiMo-V2.6的具体基准测试成绩及与同类开源全模态模型（如Qwen-VL系列）的横向对比仍有待检验65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/988881API
curl https://kongchang.com/api/v1/knowledge/claims/988881MCP
get_claim(id=988881)