SvelteBench: leaderboard
Nine component-writing tasks for Svelte 5, covering runes such as derived, effect and props plus snippets and each blocks. Each task is sampled ten times by default, once for expensive models, and each generated component is run against the task's Vitest tests. The board averages pass@1, the chance one sample passes all of a task's tests, over the nine tasks; higher is better. Source: github.com/khromov/svelte-bench.
Metric: Average pass@1 (%). Source: khromov.github.io. Saturation forecast: Estimated already saturated. 242 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 100 | #58 |
| 2 | Claude Sonnet 4.6 | 100 | #90 |
| 3 | GPT-5.4 | 100 | #79 |
| 4 | GPT-5.5 | 100 | #31 |
| 5 | Claude Opus 4.6 | 100 | #63 |
| 6 | Claude Opus 4.7 | 100 | #49 |
| 7 | Gemma 4 31B (IT) | 100 | #176 |
| 8 | Claude Opus 4.8 | 100 | #37 |
| 9 | Claude Opus 4.5 | 100 | #83 |
| 10 | GPT-5.6 Luna | 100 | #68 |
| 11 | Claude Sonnet 5 | 100 | #57 |
| 12 | GPT-6 Astra | 100 | #6 |
| 13 | Gemini 3.8 Flash | 100 | #21 |
| 14 | DeepSeek V4.1 Flash | 100 | #36 |
| 15 | Grok 4.5 | 100 | #48 |
No result here: #9 Claude Fable 5.1, #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=sveltebench · How It Works · Data refreshed daily, snapshot 2026-10-20.