AI Agent Evaluations for Next.js (Next.js Evals)
We show this table for reference; we do not rank on it.
A Vercel benchmark for AI coding agents on Next.js code generation and migration tasks, reporting pass@4, average execution time, and results with bundled documentation through AGENTS.md.
Success rate (pass@4) on Next.js Evals — November 2, 2026
We mirror the published success rate (pass@4) view for Next.js Evals. Claude Fable 5.1 (high) leads the public snapshot at 100%, followed by Claude Opus 5.5 (high) (100%) and Gemini 4 Argon (100%). We do not use these results to rank models overall.
Claude Fable 5.1 (high)
Anthropic
Claude Code
Claude Opus 5.5 (high)
Anthropic
Claude Code
Gemini 4 Argon
OpenCode
16 modelsCodingCurrentDisplay onlyUpdated November 2, 2026
Success rate (pass@4) table (16 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Fable 5.1 (high)AnthropicClaude Code | 100% | Not reported | Unknown |
| 2 | Claude Opus 5.5 (high)AnthropicClaude Code | 100% | Not reported | Unknown |
| 3 | Gemini 4 ArgonGoogleOpenCode | 100% | Not reported | Closed |
| 4 | GPT 6 Sol (high)OpenAICodex | 100% | Not reported | Unknown |
| 5 | GPT 6.1 Terra (high)OpenAICodex | 100% | Not reported | Unknown |
| 6 | Grok 4.7xAIOpenCode | 96% | Not reported | Closed |
| 7 | Gemini 3.8 FlashGoogleOpenCode | 93% | Not reported | Closed |
| 8 | GPT 6 Astra (high)OpenAICodex | 93% | Not reported | Unknown |
| 9 | Pixel CanaryUnknownOpenCode | 93% | Not reported | Unknown |
| 10 | GPT 6 Luna (high)OpenAICodex | 86% | Not reported | Unknown |
| 11 | Kimi K3Moonshot AIOpenCode | 86% | Not reported | Open |
| 12 | Claude Sonnet 5AnthropicClaude Code | 82% | Not reported | Closed |
| 13 | Composer 2.5CursorCursor | 82% | Not reported | Closed |
| 14 | GLM 5.2Z.AIOpenCode | 82% | Not reported | Unknown |
| 15 | MiniMax M3MiniMaxOpenCode | 82% | Not reported | Open |
| 16 | Kimi K2.7 CodeMoonshot AIOpenCode | 75% | Not reported | Open |
How Next.js Evals is shown here
We mirror Vercel's current Next.js agent table from the November 2, 2026 export. Each row reports pass@4 across 28 evals: an eval passes when any of four attempts succeeds. The AGENTS.md column shows results with bundled Next.js documentation.
This snapshot contains 16 current agents. Vercel lists 22 superseded agents separately with their last measured dates; those agents are not rerun. Newly added evals that a model has never run count against its success rate; reworked evals retain that model's last result until rerun. Infrastructure failures are discarded and rerun.
Each row combines a model with an agent harness. We keep these framework-specific results outside weighted model rankings.
Snapshot
The published Next.js Evals snapshot places Claude Fable 5.1 (high) first at 100%. The third row is 0.0 points behind. The broader top-10 range is 14.0 points, so the table still separates the published systems.
16 models have been evaluated on Next.js Evals. The benchmark falls in the Coding category. Next.js Evals is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About Next.js Evals
Year
2026
Tasks
31 current Next.js code generation and migration tasks
Format
Success rate (pass@4), with and without bundled documentation
Difficulty
Framework-specific web application engineering
The September 25, 2026 export contains 14 current agents across 31 evals. An eval passes when any of four attempts succeeds. Vercel lists 21 superseded agents separately without rerunning them. Newly added evals that a model has never run count against its success rate; reworked evals retain its last result until rerun. Rows combine model choice with an agent harness and remain outside weighted rankings.
Freshness and provenance
Version
Next.js Evals 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public tasks with withheld Vitest assertions
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does Next.js Evals measure?
A Vercel benchmark for AI coding agents on Next.js code generation and migration tasks, reporting pass@4, average execution time, and results with bundled documentation through AGENTS.md.
Which model leads the published Next.js Evals snapshot?
Claude Fable 5.1 (high) currently leads the published Next.js Evals snapshot with 100% success rate (pass@4). BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on Next.js Evals?
The November 2, 2026 snapshot contains 16 AI models.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.