| Rank | Model | Provider | Audit Date | Overall | Reasoning | Math | Coding | Multilingual | Moroccan | Knowledge | Status |
|---|
ATU 1.1 — Agentic Tool Use
Second Compiwer AI benchmark: ATU-1.1 measures how reliably a model plans, calls tools, reads tool output, and completes multi-step agentic tasks. Scores are the official audited ATU-1.1 values.
| Rank | Model | Provider | Type | ATU-1.1 Score | Status |
|---|
Task Explorer
Sample tasks and deterministic evaluation methods per capability.
Developer Quickstart
Run local or API evaluations directly with the CLI.
CLI Evaluation
pip install -e . rmcmmk run --model Qwen/Qwen2.5-7B-Instruct --output results.json
LoRA / Fine-Tuned Checkpoints
rmcmmk run \ --model meta-llama/Meta-Llama-3-8B-Instruct \ --lora ./my-darija-lora \ --load-in-4bit
Support This Project
Help keep Compiwer AI Benchmarks (RMCMMK-Bench 1.0 + ATU 1.1) free, independent, and transparent. No paywalls, no data selling.
₿ Crypto Donations (Direct)
Direct wallet transfers — no middleman, no account needed. Send to any address below.
BTC address coming soon. ETH/USDT and SOL ready now.
EVM address works on: Ethereum, Linea, Base, Arbitrum, BNB, OP, Polygon, Monad
💼 Hire Me (Private Benchmarks)
Need a private model evaluation? I run confidential benchmarks for companies and researchers. Paid in crypto, bank transfer, or your preferred method.
📺 Content & Tutorials
Watch and share benchmark analysis, model reviews, and Moroccan AI research. YouTube ad revenue requires no PayPal.
Follow on X for updates🐦 Brave Browser Tips
Use the Brave browser and tip this site in BAT (Basic Attention Token). No account needed beyond your crypto wallet.
- ⭐ Star the GitHub repo
- 🐦 Share benchmark results on X with credit
- 🤗 Upvote on Hugging Face
- 📝 Submit benchmark tasks via GitHub Issues
- 🔗 Link to rmcmmk.vercel.app from your project
📬 Get Notified
New benchmarks, model releases, and updates. No spam, unsubscribe anytime.