Odysseus AI Hardware Requirements

Updated June 2026

The hardware you need depends entirely on what models you want to run locally. Odysseus itself is lightweight - it's the LLM inference that needs power. You can also skip local hardware entirely by connecting to cloud APIs (OpenRouter, OpenAI).

VRAM Tiers

CPU Only

VRAM: No GPURAM: 8GB+Storage: 10GB

Slow inference (2-5 tokens/sec). Usable for simple tasks. Not recommended for daily use.

TinyLlama 1.1BPhi-2 2.7BGemma 2B

Entry GPU

VRAM: 8GBRAM: 16GBStorage: 20GB

Good for personal chat and basic code assistance. 10-20 tokens/sec depending on model.

Llama 3.1 8B (Q4)Mistral 7B (Q4)Gemma 2 9B (Q4)CodeLlama 7B

Mid-Range GPU

Recommended

VRAM: 12GBRAM: 16GB+Storage: 40GB

Comfortable for most tasks. Good balance of speed and model quality. This is the sweet spot.

Llama 3.1 8B (FP16)CodeLlama 13B (Q4)Qwen2 14B (Q4)Phi-3 14B (Q4)

High-End GPU

VRAM: 24GB+RAM: 32GB+Storage: 100GB+

Full capability. Can run the largest quantized models with good performance.

Llama 3.1 70B (Q4)Mixtral 8x7BDeepSeek V2Command R+ (Q4)

GPU Quick Reference

VRAM	NVIDIA	AMD	Apple
8GB	RTX 3060 Ti / 4060	RX 7600	M1/M2 (shared)
12GB	RTX 3060 12GB / 4070	RX 7700 XT	M1 Pro 16GB
16GB	RTX 4080 / A4000	RX 7800 XT	M2 Pro 16GB
24GB	RTX 3090 / 4090 / A5000	RX 7900 XTX	M2 Max 32GB

No GPU? No Problem.

You can connect Odysseus to cloud APIs (OpenRouter, OpenAI) for inference. This means you don't need any GPU at all - just a working internet connection. The tradeoff is cost (per-token pricing) and privacy (your prompts go through a third-party service).