Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Coding benchmark
ProgramBench: Can Language Models Rebuild Programs From Scratch?. Every model the catalog carries a published ProgramBench value for, ranked by that value.
A cleanroom software-engineering benchmark where agents receive only a compiled executable and documentation, then must architect and implement a complete codebase that reproduces the original program's behavior. 21 models in the ModelChoice catalog have a published value.
Each point is a model with a published ProgramBench value and complete token prices, on the 75% input / 25% output blend. Frontier models are spaced evenly along the cost axis.
21 models with a published ProgramBench value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.
| Rank | Model | Provider | Cleanroom executable reimplementation |
|---|---|---|---|
| 1 | Anthropic | 93.0 | |
| 2 | Anthropic | 91.2 | |
| 3 | Anthropic | 87.6 |
What the benchmark measures and how its value is published, from the source’s own record.
ProgramBench (ProgramBench: Can Language Models Rebuild Programs From Scratch?) is a coding benchmark published by ProgramBench: Can Language Models Rebuild Programs From Scratch. A cleanroom software-engineering benchmark where agents receive only a compiled executable and documentation, then must architect and implement a complete codebase that reproduces the original program's behavior. Its evaluation set is described by the source as 200 program reconstruction tasks. The source rates its difficulty as full-repository software architecture.
Each model's result is published by ProgramBench: Can Language Models Rebuild Programs From Scratch as cleanroom executable reimplementation. ModelChoice prints that value to 1 decimal place, exactly as published, and never converts or rescales it. This page ranks the 21 models that carry a published value, highest value first; a model the source never scored is left out rather than ranked at zero. The source does not state whether a higher value is the better result, so neither does this page.
| 4 | Anthropic | 85.6 |
| 5 | StepFun | 80.5 |
| 6 | Anthropic | 79.7 |
| 7 | Moonshot AI | 77.8 |
| 8 | Z.AI | 63.7 |
| 9 | Meta | 59.0 |
| 10 | Moonshot AI | 53.6 |
| 11 | Meta | 50.0 |
| 12 | Alibaba | 31.5 |
| 13 | Tencent | 30.0 |
| 14 | Baidu | 29.5 |
| 15 | DeepSeek | 29.1 |
| 16 | Xiaomi | 26.5 |
| 17 | Xiaomi | 26.0 |
| 18 | Meituan | 25.5 |
| 19 | Z.AI | 19.0 |
| 20 | Tencent | 17.5 |
| 21 | Tencent | 16.0 |
Evidence key: Observed
Rows are ordered by the value ProgramBench: Can Language Models Rebuild Programs From Scratch? published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.