flash-1-mini is here — free & Apache 2.0
AI you can own.
Open-weight models that run on your computer.
Your data never leaves. No API key. No subscription.
Verified, not promised
more reliable legal citations
flash-1-mini gets Canadian legal citations right 42.1% of the time — versus 15.8% for the Qwen base it's fine-tuned from, measured under identical conditions on the CBLRE benchmark.
It also follows multi-part instructions 22.9 points more accurately (IFEval). The capabilities it trades away are published right next to the gains — no cherry-picking.
See the full benchmarksNewsroom
Latest from SimpleDirect
Evaluating AI for Canadian regulated work: a methodology
A vendor-neutral protocol for measuring how AI models perform on Canadian legal, privacy, and constitutional reasoning — bilingual, reproducible, and citable in procurement.
Introducing CBLRE: a public benchmark for Canadian bilingual legal AI
The Canadian Bilingual Legal & Regulatory Evaluation is open — six tracks, bilingual ground truth, reproducible scoring. No model scores in the dataset card; those come separately, after expert validation.
How we measure our models: the benchmarking protocol
The full evaluation suite, serving configuration, scoring rules, and reporting standards we apply to every SimpleDirect model — built for one thing: every number we publish must be reproducible.
Ownership
What ownership actually means
You have the weights.
The model files live on your machine. They are files, not a service.
It runs without internet.
Once downloaded, inference happens entirely on your hardware. Airplane mode works.
Nobody can revoke access.
No account to suspend. No API key to rotate. If SimpleDirect disappeared tomorrow, your model still works.
Your data stays local.
Every prompt, every document, every conversation stays on your hardware. Privacy is architectural, not a setting.
No per-token billing.
You downloaded it once, for free. Run it a million times. The cost is your electricity.
No vendor lock-in.
Models ship as GGUF files. Compatible with Ollama, LM Studio, llama.cpp, and any compliant runtime.
The models
Three models. One principle: yours.
flash-1-mini
Available now- Size
- 4 billion parameters
- Download
- ~2.7 GB download
- Runs on
- Any laptop (4 GB RAM)
- Best for
- Personal AI, quick tasks, mobile
Free — forever
flash-1
Coming September 30, 2026- Size
- 9 billion parameters
- Download
- ~5.5 GB download
- Runs on
- 8 GB RAM or entry GPU
- Best for
- Business workloads, RAG, daily driver
Free — forever
flash-1-pro
Coming March 31, 2027- Size
- 27 billion parameters
- Download
- ~16 GB download
- Runs on
- 24+ GB VRAM or 32+ GB RAM
- Best for
- Enterprise, defense, complex reasoning
Free — forever
All models ship as GGUF with multiple quantization levels. Bilingual English / French. Citation-grounded responses.
Released alongside the model
Four public goods
We're opening the evaluation infrastructure too — so anyone can reproduce and audit the numbers.
Canadian Bilingual Legal Corpus
The open dataset for Canadian-context AI.
CBLRE Evaluation Suite
Six tracks, bilingual ground truth, reproducible scoring.
Canadian AI Evaluation Methodology v1.0
How to evaluate AI for Canadian regulated workflows.
Model Benchmarking Methodology v1.0
The reproducibility protocol.
Where it fits
You already have the hardware.
Now you need the brain.
SimpleDirect models run at every level of the ownership ladder — from the laptop on your desk to the rack in your colo. Same models. Same weights. Your choice of home.
Tier 1
Your laptop
- Hardware
- Mac Mini, MacBook, any laptop with 4+ GB RAM
- Model
- flash-1-mini (free)
- Good for
- Solo founders, consultants, creators
Your first private AI. Zero cost after the download.
Tier 2
Your office
- Hardware
- Mac Studio, workstation with 32–96+ GB RAM
- Model
- flash-1 or flash-1-pro
- Good for
- Law firms, accounting practices, clinics, small teams
Serious capability on your desk. Power draw: three light bulbs.
Tier 3
Your cloud
- Hardware
- Bare-metal GPU in a datacenter you choose
- Model
- flash-1-pro + vLLM for multi-user
- Good for
- Growing businesses, regulated industries, multi-office teams
Pick the jurisdiction. Own the stack. Move providers in a weekend.
Tier 4
Your rack
- Hardware
- NVIDIA GPU in a colocation facility
- Model
- flash-1-pro + deployment support
- Good for
- Enterprise, defense, critical operations
Full ownership. Air-gapped if needed. No external dependencies. Ever.
Not sure which level you need?
Read the full ownership spectrumWho it's for
Built for people who handle
things that matter
Founders
You're paying $20–200/month for AI that isn't yours. Download flash-1-mini and own it in three minutes.
Lawyers
Client privilege doesn't survive a ChatGPT prompt. Run your AI locally. No data leaves your machine.
Accountants & advisors
Client financial data on someone else's server is a liability. This is the fix.
Clinics & healthcare
PIPEDA, PHIPA, HIPAA — pick your regulation. Local inference solves all of them.
CTOs & engineering
Open-weight GGUF models. RAG-optimized. Function calling. Deploy on your hardware, air-gapped if needed.
Defense & government
Models that run with no external dependencies. Commercial licensing and deployment support available.
The showcase — September 30, 2026
Try the models, free.
chat.getsimpledirect.com is a free way to try SimpleDirect's open Canadian Business + Legal + Regulatory models in your browser — on Canadian sovereign compute.
No install. Just a browser.
Create a free account and start asking. Generous quotas, no payment, nothing to download.
Canadian sovereign compute.
Hosted on Canadian sovereign AI infrastructure in Rimouski, Quebec — your queries are processed under Canadian jurisdiction.
Plain-language data handling.
The transparency page spells out exactly what is and isn't stored, and what the showcase is and isn't for.
Be first when the showcase opens.
One email when chat.getsimpledirect.com goes live. No marketing blasts.
