vibecodingnav
中/EN
模型/gpt-5-6-sol

GPT 5.6 Sol

Android Bench 1.0 第三,90.8%,跑分成本 $150.0。DeepSWE v1.1 Best(max)第 4,72.7%。

OpenAI · 闭源 · 数据抓取 2026-10-10

LMArena WebDev
1616
#31 / 153 · ±6
LMArena Image-to-WebDev
1609
#12 / 65 · ±9
Terminal-Bench 4.0
37.3%
#22 / 35 · ±3.8

基准成绩

超过 90 天未更新的榜单不计入推荐
基准设置分数排名成本数据日期
LMArena WebDevxhigh1616±6#31—2026-10-27
LMArena Image-to-WebDevxhigh1609±9#12—2026-10-19
Terminal-Bench 4.0max · Codex37.3±3.8#22$2.5k2026-10-06
The Slop Index—69#11—2026-10-28
EQ-Bench Creative Writing v3—1973.5#12—2026-10-29
Convex Evals2 runs84.82#7$2.522026-10-28
ReactBenchmax46.741.5–51.8#4$3.062026-10-19
FrontierCodemax · codex27.6#10$5.502026-10-27
DeepSWE v1.1max · mini-swe-agent72.769.8–75.5#4$8.392026-09-22
SWE-bench Verified—榜单已停更——2026-02-26
Aider Polyglot—榜单已停更——2025-11-20

选不选它

为什么选
  • 标准任务前三,和 Anthropic 前两名几乎打平。
什么时候别选
  • 成本 $150,和榜首同一量级。

出现在这些场景

由场景榜单自动汇总
做 Web 前端#4DeepSWE 72.7%做一个 Android App#4DeepSWE 72.7%写 CLI、脚本与 DevOps#4DeepSWE 72.7%写文档和文案#11Slop Index 69

同级对比