README.md

---
title: Python → Rust & C++
emoji: ⚡
colorFrom: purple
colorTo: pink
sdk: gradio
sdk_version: 6.23.1
app_file: app.py
pinned: false
short_description: LLM ports Python to Rust or C++, measured and verified
---

# Python → Rust & C++

Give it a Python program. An LLM rewrites it in **Rust or C++**. Both are then
compiled and executed under the same sandbox, so the speedup is **measured, not
estimated** — and the two outputs are compared, because a translation that is
fast but wrong is not a translation.

## Measured results

Max subarray sum, O(n²), n=10 000 × 20 seeds, measured on a laptop:

| Target | Model | Python | Compiled | Speedup | Output |
|---|---|---:|---:|---:|:--:|
| C++ | GPT-5 Nano | 24.739 s | 0.337 s | **73×** | identical ✓ |
| Rust | GPT-5 | 24.518 s | 0.391 s | **63×** | identical ✓ |
| Rust | GPT-5 Nano | 24.590 s | 0.405 s | **61×** | identical ✓ |

Timings on this Space will be slower — it runs on shared CPU. The *ratio* is the
point and stays representative.

## Notes

- **Free to use.** The demo runs on a free-tier model. Frontier models (GPT-5,
  Claude, Grok) require your own API key — pasted keys are used for a single
  request and never logged, stored, or written to disk.
- **Submitted code is sandboxed.** It runs in a separate process with an
  allowlisted environment containing no API keys, a wall-clock timeout, and
  CPU, memory, and process-count limits.
- **First load may take ~30 seconds** if the Space has been idle.

Source and full write-up: <https://github.com/Yugjohri/python-to-rust-cpp>