# Web Clone Bench v1 (Web Clone Bench)

> Coding agents recreate nine reference web apps from an empty directory. Plato checks visual and functional behavior through end-to-end workflows.

Canonical page: https://benchlm.ai/benchmarks/web-clone-bench

- Category: [Coding](/coding)
- Last updated: October 15, 2026 snapshot

We mirror Plato’s Web Clone Bench v1 tables: 29 effort configurations across 7 models and 9 web apps. Agents explore a running reference and rebuild it from an empty directory. End-to-end workflows check both visual and functional behavior.

Each row includes the vendor’s coding harness and reasoning-effort choice. The table keeps the published full score, average cost and time per run, steps, and per-app scores together. These results remain outside weighted model rankings.

- [Plato method and result tables](https://plato.so/research/web-clone-bench/)

## About Web Clone Bench

- Year: 2026
- Tasks: 9 reference web apps
- Format: Published full score (%)
- Difficulty: Long-running web application recreation
- Paper: [Introducing Web Clone Bench v1](https://plato.so/research/web-clone-bench/)

Each agent explores a running reference app and builds its own replica without starting code. Runs use each vendor's coding harness and end at submission or an eight-hour limit. We preserve all 29 published model-and-effort configurations across seven models, including average run cost, time, steps, and scores for each app.

Recreating a page is one frontend task. The [app-development model shortlist](/best/frontend-app-dev) keeps that evidence alongside other checks for the job you need.

Web Clone Bench is currently displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Leaderboard (29 effort configurations)

| Rank | Model | Effort | Full score | Avg cost / run (USD) | Avg time / run | Steps |
|---|---|---|---|---|---|---|
| 1 | [Claude Opus 5.5](/models/claude-opus-5-5) | xhigh | 34.2% | $652.07 | 5h 36m | 10,305 |
| 2 | [Claude Opus 5.5](/models/claude-opus-5-5) | max | 32.3% | $718.71 | 5h 46m | 10,571 |
| 3 | [Claude Opus 5.5](/models/claude-opus-5-5) | high | 30.5% | $546.59 | 5h 57m | 10,302 |
| 4 | [Claude Opus 5.5](/models/claude-opus-5-5) | medium | 18.6% | $403.88 | 4h 4m | 6,609 |
| 5 | [GPT-6 Astra](/models/gpt-6-astra) | max | 17.5% | $207.27 | 6h 20m | 1,290 |
| 6 | [Claude Fable 5.1](/models/claude-fable-5-1) | high | 13.3% | $272.24 | 2h 13m | 2,768 |
| 7 | [Claude Fable 5.1](/models/claude-fable-5-1) | xhigh | 13.0% | $314.08 | 2h 32m | 2,807 |
| 8 | [GPT-6.1 Sol](/models/gpt-6-1-sol) | max | 12.6% | $25.35 | 6h 40m | 793 |
| 9 | [GPT-6 Astra](/models/gpt-6-astra) | xhigh | 11.8% | $116.64 | 2h 39m | 585 |
| 10 | [Claude Fable 5.1](/models/claude-fable-5-1) | max | 11.6% | $227.84 | 1h 51m | 2,565 |
| 11 | [GPT-6 Astra](/models/gpt-6-astra) | high | 10.0% | $75.24 | 1h 42m | 434 |
| 12 | [GPT-6.1 Sol](/models/gpt-6-1-sol) | high | 9.9% | $14.25 | 2h 45m | 628 |
| 13 | [GPT-6 Astra](/models/gpt-6-astra) | medium | 9.6% | $45.81 | 57m | 278 |
| 14 | [Claude Fable 5.1](/models/claude-fable-5-1) | low | 9.2% | $174.75 | 2h 7m | 2,311 |
| 15 | [Claude Fable 5.1](/models/claude-fable-5-1) | medium | 9.1% | $206.55 | 2h 9m | 2,188 |
| 16 | [GPT-6.1 Sol](/models/gpt-6-1-sol) | xhigh | 8.6% | $16.78 | 3h 15m | 672 |
| 17 | [GPT-6 Astra](/models/gpt-6-astra) | low | 7.8% | $30.69 | 35m | 189 |
| 18 | [GPT-6.1 Sol](/models/gpt-6-1-sol) | medium | 6.9% | $5.80 | 1h 2m | 271 |
| 19 | [GPT-6.1 Sol](/models/gpt-6-1-sol) | low | 4.6% | $2.49 | 41m | 209 |
| 20 | [Grok 4.7](/models/grok-4-7) | low | 2.8% | $18.32 | 26m | 532 |
| 21 | [Grok 4.7](/models/grok-4-7) | high | 2.7% | $19.22 | 33m | 527 |
| 22 | [Claude Opus 5.5](/models/claude-opus-5-5) | low | 2.4% | $16.95 | 25m | 496 |
| 23 | [Grok 4.7](/models/grok-4-7) | medium | 2.1% | $15.91 | 32m | 437 |
| 24 | [Muse Spark 1.3](/models/muse-spark-1-3) | medium | 1.5% | $5.46 | 32m | 235 |
| 25 | [Gemini 3.8 Flash](/models/gemini-3-8-flash) | medium | 1.0% | $2.01 | 17m | 219 |
| 26 | [Muse Spark 1.3](/models/muse-spark-1-3) | low | 0.9% | $1.89 | 16m | 98 |
| 27 | [Gemini 3.8 Flash](/models/gemini-3-8-flash) | high | 0.5% | $2.72 | 23m | 300 |
| 28 | [Gemini 3.8 Flash](/models/gemini-3-8-flash) | low | 0.2% | $1.66 | 11m | 227 |
| 29 | [Muse Spark 1.3](/models/muse-spark-1-3) | high | 0.2% | $4.20 | 26m | 179 |

Each model uses its vendor's coding harness. Costs are USD per run. Full score and steps retain the source labels; the exact score formula and repeat counts are undisclosed.

### Full scores by app

| Model | Effort | Gmail | Figma | Sheets | Slack | DocuSign | Jira | Shopify | Craigslist | Zendesk |
|---|---|---|---|---|---|---|---|---|---|---|
| [Claude Opus 5.5](/models/claude-opus-5-5) | xhigh | 65.0% | 14.0% | 32.5% | 36.7% | 51.2% | 8.6% | 41.8% | 18.4% | 39.7% |
| [Claude Opus 5.5](/models/claude-opus-5-5) | max | 65.8% | 14.2% | 18.4% | 34.0% | 48.9% | 20.4% | 13.1% | 24.5% | 51.8% |
| [Claude Opus 5.5](/models/claude-opus-5-5) | high | 71.8% | 7.6% | 27.4% | 20.3% | 41.5% | 9.3% | 21.9% | 20.8% | 53.5% |
| [Claude Opus 5.5](/models/claude-opus-5-5) | medium | 42.3% | 13.3% | 21.9% | 13.0% | 22.1% | 5.2% | 13.6% | 18.6% | 17.4% |
| [GPT-6 Astra](/models/gpt-6-astra) | max | 33.8% | 4.6% | 8.2% | 13.9% | 8.7% | 9.2% | 38.7% | 7.8% | 32.6% |
| [Claude Fable 5.1](/models/claude-fable-5-1) | high | 57.6% | 2.2% | 2.6% | 7.6% | 9.2% | 6.3% | 11.5% | 14.7% | 7.7% |
| [Claude Fable 5.1](/models/claude-fable-5-1) | xhigh | 28.2% | 2.6% | 6.3% | 7.6% | 7.1% | 6.8% | 22.1% | 20.9% | 15.3% |
| [GPT-6.1 Sol](/models/gpt-6-1-sol) | max | 30.4% | 3.3% | 7.1% | 12.2% | 13.2% | 5.4% | 16.6% | 9.8% | 15.1% |
| [GPT-6 Astra](/models/gpt-6-astra) | xhigh | 36.1% | 1.9% | 4.7% | 11.7% | 7.9% | 6.8% | 16.3% | 10.4% | 10.4% |
| [Claude Fable 5.1](/models/claude-fable-5-1) | max | 34.7% | 2.1% | 4.2% | 7.3% | 5.8% | 4.0% | 14.1% | 19.6% | 12.3% |
| [GPT-6 Astra](/models/gpt-6-astra) | high | 11.0% | 2.7% | 6.7% | 10.1% | 14.4% | 4.7% | 13.0% | 10.7% | 16.3% |
| [GPT-6.1 Sol](/models/gpt-6-1-sol) | high | 25.9% | 0.7% | 4.4% | 13.6% | 6.2% | 6.3% | 12.0% | 11.0% | 9.1% |
| [GPT-6 Astra](/models/gpt-6-astra) | medium | 17.3% | 3.8% | 6.3% | 12.0% | 6.6% | 5.3% | 11.3% | 8.8% | 15.1% |
| [Claude Fable 5.1](/models/claude-fable-5-1) | low | 28.9% | 1.5% | 4.9% | 11.7% | 4.5% | 2.0% | 5.7% | 15.6% | 8.4% |
| [Claude Fable 5.1](/models/claude-fable-5-1) | medium | 18.0% | 2.0% | 5.4% | 7.2% | 6.6% | 5.2% | 13.1% | 20.9% | 3.7% |
| [GPT-6.1 Sol](/models/gpt-6-1-sol) | xhigh | 15.6% | 6.2% | 5.1% | 7.2% | 3.3% | 6.9% | 9.0% | 11.3% | 12.6% |
| [GPT-6 Astra](/models/gpt-6-astra) | low | 23.9% | 1.2% | 4.9% | 8.2% | 4.1% | 4.0% | 7.8% | 2.1% | 14.0% |
| [GPT-6.1 Sol](/models/gpt-6-1-sol) | medium | 14.4% | 3.7% | 3.0% | 13.2% | 3.3% | 4.5% | 8.9% | 3.0% | 8.0% |
| [GPT-6.1 Sol](/models/gpt-6-1-sol) | low | 5.1% | 1.5% | 3.0% | 8.8% | 3.3% | 3.0% | 1.6% | 7.5% | 7.2% |
| [Grok 4.7](/models/grok-4-7) | low | 12.4% | 0.0% | 0.7% | 5.3% | 0.4% | 0.9% | 2.3% | 0.8% | 2.6% |
| [Grok 4.7](/models/grok-4-7) | high | 11.8% | 0.2% | 1.9% | 4.6% | 0.2% | 1.4% | 1.9% | 0.8% | 1.4% |
| [Claude Opus 5.5](/models/claude-opus-5-5) | low | 7.7% | 0.0% | 1.7% | 5.8% | 0.2% | 1.3% | 0.0% | 0.0% | 4.5% |
| [Grok 4.7](/models/grok-4-7) | medium | 9.6% | 0.4% | 0.0% | 4.8% | 0.2% | 0.0% | 3.0% | 0.0% | 1.1% |
| [Muse Spark 1.3](/models/muse-spark-1-3) | medium | 8.7% | 0.0% | 0.0% | 4.9% | 0.4% | 0.0% | 0.0% | 0.0% | 0.0% |
| [Gemini 3.8 Flash](/models/gemini-3-8-flash) | medium | 2.4% | 0.0% | 0.4% | 5.2% | 0.4% | 0.0% | 1.1% | 0.0% | 0.0% |
| [Muse Spark 1.3](/models/muse-spark-1-3) | low | 0.9% | 0.0% | 0.4% | 4.9% | 0.2% | 0.0% | 0.0% | 0.0% | 1.7% |
| [Gemini 3.8 Flash](/models/gemini-3-8-flash) | high | 2.4% | 0.2% | 0.4% | 0.0% | 0.4% | 0.0% | 0.0% | 0.0% | 1.1% |
| [Gemini 3.8 Flash](/models/gemini-3-8-flash) | low | 1.1% | 0.0% | 0.0% | 1.1% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% |
| [Muse Spark 1.3](/models/muse-spark-1-3) | high | 1.8% | 0.0% | 0.0% | 0.0% | 0.2% | 0.0% | 0.0% | 0.0% | 0.0% |

## FAQ

### What does Web Clone Bench measure?

Coding agents recreate nine reference web apps from an empty directory. Plato checks visual and functional behavior through end-to-end workflows.

### Which model leads the published Web Clone Bench snapshot?

Claude Opus 5.5 (xhigh) currently leads the published Web Clone Bench snapshot with a score of 34.2%.

### How many models are evaluated on Web Clone Bench?

The October 15, 2026 snapshot contains 29 AI models.

### Does Web Clone Bench affect BenchLM's overall score?

Not directly. Web Clone Bench is still displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.