当前排名

编程

截至 2026-09-13 21:17:03 UTC，近 7 日编程体感排名：闭源组 — Claude Fable 5.1（74.9 分，n=28）；开源组 — Qwen3.8 27B（74.8 分，n=36）。

每个场景展示 Top 3 具体版本，样本量满 5 条即可上榜；排名变化对比上一次发布的榜单。

闭源模型开源模型

#1 Claude Fable 5.1n=28 74.9▾

“I still prefer fable for architecture design, multi file codebase changes, etc. at least for me, in the domain of web apps, fable feels more reliable and writes cleaner code.”
来自 Reddit · 英文 · 2026-09-13

“Fable 5.1 beats astra in code quality, but astra does almost everything else better.”
来自 Reddit · 英文 · 2026-09-13

查看模型详情 →

#2 Claude Fable 5n=62 67.9▾

“Solving obscure problems by writing code, usually Fable but with a smaller margin”
来自 Reddit · 英文 · 2026-09-13

“Fable was for more significant load bearing ones like new architecture.”
来自 Reddit · 英文 · 2026-09-13

查看模型详情 →

#3 Claude Sonnet 5n=5 66.5→

#1 Qwen3.8 27Bn=36 74.8▾

“Glimmer is not better than qwen3.8:27b for coding by a very very big margin”
来自 Reddit · 英文 · 2026-09-12

“My local model spits out code and tests 24/7”
来自 Reddit · 英文 · 2026-09-12

查看模型详情 →

#2 GLM 5.3 Flashn=27 65.3▾

“在代码这块，glm5.3flash我没感到有什么特殊之处”
来自知乎 · 2026-09-13

“GLM 5.3 Flash (Excellent+ all-rounder)”
来自 Reddit · 英文 · 2026-09-12

查看模型详情 →

#3 GLM 5.3n=18 63.8▾

“If Luna MAX works for you, 5.3 probably will too since it's the stronger model.”
来自 Reddit · 英文 · 2026-09-13

“GLM-5.3 is a very capable software engineering model.”
来自 Reddit · 英文 · 2026-09-12

查看模型详情 →