Strict Programming Language

Strict Programming Language

  • Games
  • Performance
  • VS Code
  • Docs
  • Blog
  • GitHub

›Source

Documentation

  • Overview
  • The VS Code Experience
  • Getting Started
  • Platforms
  • Built with Strict
  • Cheat Sheet

Language

  • Types
  • Members and Variables
  • Decimals and Exact Values
  • Methods
  • Tests
  • Statements and Expressions
  • If
  • For
  • Lists and Dictionaries
  • Errors
  • Keywords
  • Grammar

Rules

  • Coding Style
  • Code Validations
  • Naming

Source

  • Base Types
  • Architecture
  • Examples
  • Machine Learning
  • Benchmarks

More

  • Using APIs
  • Concurrency
  • Microservices

Benchmarks

Where the code behind each chart in the Performance section lives and how to run it. Everything is written in Strict and builds with strict build, comparisons use the same inputs and settings.

All paths are in the Strict repository unless noted.

BenchmarkSourceCompared against
Image BlurImageProcessing/BenchBlur.strictImageProcessing/Compare: C, Rust, Goose, C#, TypeScript, Python, NumPy
Qwen3 modelsNeuralNetworks/Qwen3Benchmark.strictllama.cpp, Ollama
Qwen3.8 and reasoning modelsNeuralNetworks/Qwen3Benchmark.strictllama.cpp, Ollama
GPT-2 text generationNeuralNetworks/Gpt2Benchmark.strictPyTorch with Hugging Face transformers
Moss image generationStrictComfy MossImage.strict, coming soonComfyUI
MiniMax H3 videoStrictComfy H3Video.strict, coming soonComfyUI

Image Blur

A 9x9 box blur of a 4096x2048 photo, written as a plain loop. Everything needed (input image, build and run commands for every language) is described in the Compare README.

strict build ImageProcessing/BenchBlur.strict
ImageProcessing/BenchBlur.exe

Set STRICT_THREADS=1 for one thread or STRICT_GPU=0 for CPU only.

Qwen3 and Qwen3.8

The model code is in Qwen3.strict and Qwen3Block.strict, the GGUF reader in Gguf.strict. Qwen3Benchmark.strict reports prefill and decode speed, Qwen3Text.strict generates text. The baseline is llama.cpp (b11658, CUDA) on the same GGUF files and GPU.

strict build NeuralNetworks/Qwen3Benchmark.strict

See Machine Learning for the downloads and measured results.

GPT-2

Gpt2.strict, Gpt2Block.strict and Gpt2Tokens.strict hold the model, Gpt2Benchmark.strict measures milliseconds per token. The baseline is PyTorch with Hugging Face transformers, best of 3 runs.

strict build NeuralNetworks/Gpt2Benchmark.strict

Moss image and MiniMax H3 video

Both run through StrictComfy, which loads the same model files and LoRAs as ComfyUI and runs the diffusion in Strict. The source and instructions are coming soon.

Last updated on 10/25/2026 by Benjamin Nitschke
← Machine LearningUsing APIs →
Docs
The VS Code ExperienceGetting StartedPlatformsBuilt with StrictLanguage GuideCheat SheetCoding Style
Source
Base TypesArchitectureExamplesQwen3 and Machine LearningC and C# Interop
More
BlogGitHubIssuesVS Code ExtensionWebsite SourceStar
Copyright © 2000-2026 Benjamin Nitschke