// cosmogenic.org — model intelligence — 2026.02

Open vs Closed
Model Landscape

// foundation models · edge viability · VRAM matrix · quantisation guide

The frontier has shifted. Open-weight models now match or exceed closed proprietary systems on most inference tasks. The question is no longer capability — it's deployment: which models run on the hardware you actually have.

Open-weight (self-hostable)
Closed / API-only
Hybrid / partial open
Edge-viable zone
// model parameter size vs edge viability · bubble size = VRAM required · click any model
Edge Viability →
Edge viable zone
← Smaller Model · Parameter Count · Larger Model →
// model catalogue — click to inspect
// VRAM requirements by model size and quantisation
VRAM Requirement Matrix

Quantisation trades a small accuracy penalty for dramatic reductions in VRAM. INT4 quantised 70B runs on a 48GB workstation GPU. A 7B INT4 model fits in 4GB — a gaming laptop. The column headers are GPUs you can actually buy.

Model Size Precision VRAM needed 4GB GPU
GTX 1650
8GB GPU
RTX 3070
16GB GPU
Arc A770
24GB GPU
RTX 3090
48GB GPU
RTX 6000 Ada
80GB
A100 SXM
1B–3B FP16 2–6GB
7B FP16 14GB ~
7B INT4 3.5GB
13B FP16 26GB
13B INT4 6.5GB
70B FP16 140GB ~×2
70B INT4 35GB
70B INT2 17GB ~
405B INT4 202GB ~×3
CPU only INT4 RAM 7B runs on any modern CPU via llama.cpp · ~5 tok/s · No GPU required · R740 viable today
// select a model card above
Model Detail
Click any card to inspect
This panel shows full capability breakdown, edge deployment requirements, quantisation options, and relevance for the cosmogenic federated inference stack.
// decision guide
Edge Deployment
Decision Tree
R740 CPU only · Right now
Llama 3.2 3B · Phi-3.5 Mini · Mistral 7B INT4 · Gemma 2 2B — runs today, no GPU needed
R740 + 5060 GPU · Current setup
Llama 3.1 8B FP16 · Mistral 7B FP16 · Qwen 2.5 14B INT4 · Phi-4 14B INT4
+ RTX 5060 Ti (next step · ~£550)
Llama 3.1 70B INT4 · Mixtral 8×7B · Qwen 2.5 72B INT4 · DeepSeek Coder 33B
Closed models (API only)
GPT-4o · Claude · Gemini Ultra — no local deployment · data leaves your infrastructure