Models/glm-5-2

GLM 5.2

The eval page does not state a weights license. This entry exists to carry the score and is filed as a closed API model. Next.js is 81% without AGENTS.md and 97% with it.

Zhipu · Proprietary · Fetched 2026-10-10

LMArena WebDev

1603

#27 / 142 · ±6

The Slop Index

#28 / 39

Next.js Agent Evals

81.0%

#9 / 14

Benchmark scores

Boards older than 90 days are not counted as recommendations

BenchmarkSettingScoreRankCostAs of

LMArena WebDevmax1603±6#27—2026-10-08

The Slop Index—52#28—2026-10-11

Next.js Agent EvalsOpenCode81.0#9$0.1522026-09-25

ReactBenchmax27.222.8–31.6#53$1.232026-10-10

DeepSWE v1.1max · mini-swe-agent43.842.1–45.5#24$3.922026-09-22

SWE-bench Verified—This board has stopped updating——2026-02-26

Aider Polyglot—This board has stopped updating——2025-11-20

Use it or not

Why

With AGENTS.md, Next.js pass@4 reaches 97%, level with the front.

When not

The baseline is still 81%. Do not treat 97% as the score without the docs.

Appears in these scenarios

Collected from scenario boards

Build a Web frontend#9Next.js 81.0%Build an Android app#24DeepSWE 43.8%Write CLI, scripts, and DevOps#24DeepSWE 43.8%Write docs & copy#28Slop Index 52