Public Benchmarks
Every number on this page is measured, verified, and reproducible. We run side-by-side against CPython and check that outputs match exactly.
Benchmark Methodology
Side-by-Side with CPython
Each workload runs on both Pyvorin and CPython 3.12 under identical hardware, OS, and Python configurations. No cherry-picked baselines.
Correctness Verification
Output hashes and numeric results are compared. If Pyvorin produces a different answer, the run is flagged and falls back to CPython.
Warm Runs & Compilation Cost
We report warm-run averages after JIT warm-up. Compilation time is tracked separately so you can model amortisation for your use case.
Honest Disclaimer
Results are workload-specific. Your mileage may vary. Speedups depend on code structure, data size, and how much time is spent in supported vs unsupported language features. We do not report best-case numbers as guarantees.
Proven Speedup Results
Download JSON Results| Workload | Speedup vs CPython | Source | Fit |
|---|---|---|---|
| ETL Filter / Map | 100× – 1,000×+ | Phase G7 | Strong fit |
| Group-By Aggregation | 1,000×+ | Phase G7 | Strong fit |
| Rolling Window | 4,000×+ | Phase G7 | Strong fit |
| Finance Kernels | 10× – 150×+ | Phase L | Strong fit |
| String Tokenisation | ~500× | Phase I | Strong fit |
| Log Parsing | 74×+ | Phase I | Strong fit |
| JSON ETL | 20× – 100× | Phase H | Strong fit |
| CSV ETL | 15× – 80× | Phase H | Strong fit |
All results measured on x86-64 Linux, Python 3.12, warm runs after JIT compilation. See honest_benchmark_results.json for raw data.
Workload Classification
High-Confidence Speedups
Workloads that exercise supported language features, spend most of their time in Python bytecode (not C extensions), and have minimal I/O blocking.
- Tight loops over lists, dicts, and numerics
- ETL transforms with filter, map, and group-by
- String scanning and tokenisation
- Finance kernels and rolling-window math
Limited or Variable Speedups
Workloads that hit unsupported language features, spend significant time in C extensions, or are bound by I/O, network, or GPU.
- Heavy NumPy / Pandas / Polars internals
- async/await and dynamic code (eval/exec)
- GPU training loops
- Network-bound microservices
ROI Calculator
Estimate potential savings based on your pipeline. All calculations are performed in your browser—no data is sent to our servers.
Your Pipeline
Estimated Impact
This is an illustrative estimate. Actual savings depend on workload-specific speedup, cold compile cost, and infrastructure factors.
Run Your Own Benchmark
The best benchmark is your own code. Join the pilot programme, run the benchmark suite against your workloads, and see exactly where Pyvorin helps.