Next.js AGENTS.md
PromptThe bundled Next.js docs in the eval. Each model's change is listed on its own, including zeros.
Vercel · nextjs.org/evals
Change from the baseline
Next.js Agent Evals · Framework eval · 2026-10-30 · pass@4 · 2026-10-30
Vercel writes the tasks and grades them. This measures a model together with Next.js, not one framework against another. The front sits above 90%, near the ceiling. Gemini 4 Argon and GPT 6.1 Terra (high) join the 97% first tier. Pixel Canary (OpenCode) is 9th on the source page, 90% baseline and 97% with AGENTS.md, cost N/A. This site has no vendor identity for that model, so it has no model page. Its rank stays the source rank (mapped rows leave a gap at #9).
| # | Model | Baseline | With AGENTS.md | Change |
|---|---|---|---|---|
| 1 | Claude Opus 5.5 high · Claude Code | 97.0 | 97.0 | 0 |
| 2 | GPT 6 Sol high · Codex | 97.0 | 97.0 | 0 |
| 3 | Gemini 4 Argon OpenCode | 97.0 | 97.0 | 0 |
| 4 | Claude Fable 5.1 high · Claude Code | 97.0 | 97.0 | 0 |
| 5 | GPT-6.1 Terra high · Codex | 97.0 | 94.0 | −3.0 |
| 6 | Grok 4.7 OpenCode | 94.0 | 94.0 | 0 |
| 7 | Gemini 3.8 Flash OpenCode | 90.0 | 97.0 | +7.0 |
| 8 | GPT-6 Astra high · Codex | 90.0 | 97.0 | +7.0 |
| 10 | Kimi K3 OpenCode | 84.0 | 97.0 | +13.0 |
| 11 | GLM 5.2 OpenCode | 81.0 | 97.0 | +16.0 |
| 12 | Claude Sonnet 5 Claude Code | 81.0 | 97.0 | +16.0 |
| 13 | Cursor Composer 2.5 Cursor | 81.0 | 94.0 | +13.0 |
| 14 | GPT 6 Luna high · Codex | 81.0 | 90.0 | +9.0 |
| 15 | MiniMax M3 OpenCode | 81.0 | 90.0 | +9.0 |
| 16 | Kimi K2.7 Code OpenCode | 74.0 | 90.0 | +16.0 |
Why this, and when not
Why On the same pass@4 set, you can see which models the docs move and which stay put.
Skip it Models already at 97% do not move. Do not treat the file as a universal bonus.