Introducing Runyard: Find the Right AI Model for Your Hardware

We built Runyard because we kept running into the same problem: you find an interesting LLM, download it, and then discover it doesn't fit in your VRAM — or barely fits but runs at 2 tokens per second. Runyard solves that before you waste the time.

Runyard's Model Radar is a hardware-aware model browser. You enter your specs — CPU, RAM, GPU, and VRAM — and it filters the entire catalog of major open-source LLMs to show only what will actually run on your machine. Each model comes with a performance tier, recommended quantization, and an expected tokens-per-second range.

The sunburst is a zoomable radial chart. Every arc is a model — organized by category in the inner rings (Chat, Code, Vision, Reasoning) and individual models in the outer arcs. Colors update live: green means comfortable fit, yellow is tight, red means it won't run on your current VRAM.

• ▸Click any inner ring to zoom into that category and see only those models.
• ▸Click a model arc to open its full detail panel — specs, quant options, and download links.
• ▸Click the center circle to zoom back out to the full catalog.
• ▸Use the filter bar to show only Chat, Code, Vision, or Reasoning models.

Below the sunburst is the Tier List — the same data in a ranked linear format. Every model is scored across three factors: VRAM headroom (40%), memory bandwidth utilization (35%), and benchmark quality (25%). The result is a composite 0–100 score that slots each model into a tier.

• ▸S Tier (75–100) — Elite fit. Fast, comfortable, high quality. Start here.
• ▸A Tier (60–74) — Great fit. Excellent real-world performance on your hardware.
• ▸B Tier (45–59) — Good fit. Usable with the right quantization or context setting.
• ▸C Tier (30–44) — Marginal. Runs but expect slower tok/s or tight VRAM.
• ▸D Tier (0–29) — Limited. Technically loads but performance will be poor.

• ▸No more guesswork — know before you download whether a model fits
• ▸Quantization guidance — see the best quant level for your exact VRAM
• ▸Performance expectations — get realistic tok/s estimates for your hardware
• ▸Comparison view — see multiple models side-by-side for your system
• ▸Use-case filtering — filter by coding, chat, reasoning, vision, and more

• ▸Anchor API — one endpoint for all major AI providers with unified pricing
• ▸AI YARD — fine-tune and build your own models in a guided workflow
• ▸Forge — tools and utilities for managing your local AI setup
• ▸Benchmark database — community-contributed real-world performance data
• ▸Ollama / LM Studio integration — one-click model install from Runyard

Runyard is free to use. We're building the infrastructure layer for the local AI ecosystem — starting with the tool that helps you figure out where to start. Go to runyard.dev, enter your specs, and find out what's possible on your hardware today.