// GLOSSARY -- AGENTIC AI

What is Open Weights Model?

1 min read Updated

An open-weights model is an AI model whose trained parameters are publicly released, allowing anyone to download, run, fine-tune, and deploy it — distinct from fully open-source models which also release training code and data.

WHY IT MATTERS

Open-weights models (Llama, Mistral, Qwen) have democratized AI. Instead of relying on API providers, developers can run models locally, customize them for specific domains, and deploy without per-token costs.

The 'open weights' distinction matters: Meta's Llama releases model weights under a license, but not the training code, data, or full methodology. This gives users deployment freedom without full reproducibility.

For agent builders, open-weights models offer: cost control (no API fees at scale), privacy (data never leaves your infrastructure), customization (fine-tune for your domain), and independence (no vendor lock-in).

FREQUENTLY ASKED QUESTIONS

What's the best open-weights model?
It changes rapidly. As of early 2026, Llama 3, Mistral, Qwen 2.5, and DeepSeek-R1 are leading options. The right choice depends on your size/quality tradeoff and hardware.
Are open-weights models as good as GPT-4?
The gap has narrowed significantly. Top open models (70B+) rival GPT-4 on many benchmarks. For specific domains with fine-tuning, they can exceed proprietary models.
What hardware do I need?
7B models: consumer GPU (8GB VRAM). 13B: 16-24GB. 70B: 2-4 A100s or quantized on fewer. Quantization (4-bit, 8-bit) dramatically reduces requirements with modest quality loss.

FURTHER READING

Enforce policies on every tool call

Intercept is the open-source MCP proxy that enforces YAML policies on AI agent tool calls. No code changes needed.

npx -y @policylayer/intercept
github.com/policylayer/intercept →
// GET IN TOUCH

Have a question or want to learn more? Send us a message.

Message sent.

We'll get back to you soon.