Another massive leap for the COOLJAPAN ecosystem.
Today we shipped SciRS2 0.3.1 — the most complete pure-Rust scientific computing and AI/ML infrastructure ever built.
2.59 million lines of Rust.
29 workspace crates.
Zero C/C++/Fortran dependencies.
This is the day Python’s scientific stack officially has a sovereign Rust replacement.
Why SciRS2 0.3.1 changes everything
For years, scientific computing meant depending on NumPy, SciPy, scikit-learn, and a maze of C/Fortran extensions.
SciRS2 ends that dependency forever.
- 100% Pure Rust (OxiBLAS + OxiFFT core)
- 10–100x performance on real workloads (SIMD + zero-copy + GPU)
- Memory safe + no system dependencies
- Runs natively and in the browser (WASM)
- Full Python interoperability via PyO3 + Julia bindings
What’s new in 0.3.1
- Full transformer stack (GPT-2, T5, Swin) + GNNs (GCN/GAT/GIN) + diffusion models
- Extended statistics: Gaussian processes, survival analysis, Bayesian networks, copulas
- Advanced signal processing: compressed sensing (OMP/ISTA), MFCC, source separation (ICA/NMF)
- Sparse linear algebra: LOBPCG, AMG, IRAM, BCSR/ELLPACK formats
- Time-series models: TFT, N-BEATS, DeepAR, VECM, EGARCH
- Graph algorithms: Louvain, Node2Vec, network flow
- New Julia interface + massive test coverage (19,700+ passing tests)
Benchmarks (real numbers)
- Element-wise ops: 14.17x faster than NumPy
- Convolution: 25.0x faster
- Bootstrap sampling: 30.0x faster
- Fractional FFT: 24.9x faster
- K-means (100K points): 1.18x faster
- Matrix multiply (1000×1000): 1.25x faster
This is the foundation
SciRS2 is now the computational backbone for the entire COOLJAPAN stack — powering tenrso, torsh, trustformers, oxify, tensorlogic, and the upcoming AI agents in mielinOS.
Try it today
→ https://github.com/cool-japan/scirs
Star the repo if you want scientific computing that is faster, safer, and truly sovereign.
More releases coming this week (OxiMedia 0.1.1 and others).
The Rust scientific revolution is no longer coming — it’s here.
— KitaSan at COOLJAPAN OÜ
March 9, 2026