Unverified50% confidenceBenchmarkExact time
GPT-6 Astra在Frontier Math Tier 4上峰值达到98%，在不开启推理、不使用思维链的情况下仍达到83%
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/4/2026
First Seen
Valid until: 1/2/2027
Sources
Related Entities
Related Claims
Unverified在DeepSuite基准上Grok 4.7得71%，GPT 5.6 Sol为72.7%，Fable 5.1为70%，Astra Max以74.1%登顶76% similarUnverified在Terminal Bench 2.1基准测试上，GPT-5.6 Sol的Ultra模式接近92%，超越竞品的88%表现75% similarUnverified在超高推理强度下 GPT-6.1 Sol 事实错误率为 4.1%，低于上一代的 4.5%，接近 Astra 的 4.0%75% similarUnverified在SWE-Bench Pro编程评测上，Claude Fable 5以80%的成绩领先GPT-5.6 Sol的64.6%73% similarUnverified四轮测试中Astra拿下三胜、动态图形打平，测评者判定GPT-6整体领先72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/972624API
curl https://kongchang.com/api/v1/knowledge/claims/972624MCP
get_claim(id=972624)