SWE-InfraBench - Passed Tests Share: leaderboard
SWE-InfraBench (2025) tests infrastructure-as-code editing: in 100 AWS CDK projects a code block is masked, the model writes it from a natural-language request, and unit tests check the CloudFormation template the project then synthesizes; one attempt per task. This board is the percentage of all unit tests passed across the tasks, which gives credit for partly correct solutions; higher is better.
Metric: Passed tests share (%): share of all unit tests across the tasks that the generated code passes, so partly correct solutions earn credit; 100 AWS CDK (Python) infrastructure-as-code tasks built from 34 open-source and custom repositories: the model writes a masked code block from a natural-language request inside a real project, and the synthesized CloudFormation template is checked by hidden unit tests; one attempt per task, temperature 0.25, Anthropic models prompted with extra XML tags; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.7 Sonnet | 53.1 | #458 |
| 2 | Claude 3.5 Sonnet (20240620) | 47.9 | #687 |
| 3 | Claude 3.5 Sonnet (20241022) | 47 | #637 |
| 4 | Gemini 2.5 Pro (Preview 03-25) | 42.5 | #236 |
| 5 | Gemini 2.5 Pro (Preview 05-06) | 41.1 | #275 |
| 6 | DeepSeek R1 | 34.9 | #473 |
| 7 | O4 Mini | 32.1 | #353 |
| 8 | O3 | 30.8 | #228 |
| 9 | GPT-4.1 | 26.4 | #495 |
| 10 | Llama 3.1 405B Instruct | 13 | #1115 |
| 11 | Llama 4 Maverick Instruct | 13 | #1133 |
| 12 | Gemini 2.0 Flash Lite | 11.5 | #1032 |
| 13 | Claude 3 Haiku | 10.7 | #2609 |
| 14 | GPT-4o Mini | 7 | #1623 |
| 15 | Qwen 2.5 72B Instruct Turbo | 2.7 | #1166 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=swe-infrabench-passed-tests-share · How It Works · Data refreshed daily, snapshot 2026-11-13.