what runs on your hardware?
Every open-weight model, ranked S–F for the hardware you already have. Detection runs locally in your browser — nothing leaves your machine.
| Model | Params | Min VRAM | Status |
|---|---|---|---|
| 32B | 22.0 GBQ4_K_M | — | |
| 8B | 6.5 GBQ4_K_M | — | |
| 22B | 14.7 GBQ4_K_M | — | |
| 7B | 6.9 GBQ4_K_M | — | |
| 32B | 21.5 GBQ4_K_M | — | |
| 70B | 43.0 GBQ4_K_M | — | |
| 8B | 7.5 GBQ4_K_M | — | |
| 111B | 61.0 GBQ4_K_M | — | |
| 35B | 22.5 GBQ4_K_M | — | |
| 104B | 57.0 GBQ4_K_M | — | |
| 1.5B | 3.0 GBQ8_0 | — | |
| 14B | 9.9 GBQ4_K_M | — | |
| 32B | 20.7 GBQ4_K_M | — | |
| 671B | 362.0 GBQ4_K_M | — | |
| 70B | 43.5 GBQ4_K_M | — | |
| 7B | 9.0 GBQ8_0 | — | |
| 8B | 7.5 GBQ4_K_M | — | |
| 671B | 362.0 GBQ4_K_M | — | |
| 671B | 362.0 GBQ4_K_M | — | |
| 671B | 420.0 GBQ4_K_M | — | |
| 123B | 67.0 GBQ4_K_M | — | |
| 24B | 17.0 GBQ4_K_M | — | |
| 47B | 26.0 GBQ4_K_M | — | |
| 8B | 6.0 GBQ4_K_M | — | |
| 10B | 8.5 GBQ4_K_M | — | |
| 7B | 6.8 GBQ4_K_M | — | |
| 27B | 17.7 GBQ4_K_M | — | |
| 2B | 4.0 GBQ8_0 | — | |
| 9B | 11.0 GBQ8_0 | — | |
| 12B | 10.5 GBQ4_K_M | — | |
| 1B | 2.0 GBQ8_0 | — | |
| 27B | 20.0 GBQ4_K_M | — | |
| 4B | 5.0 GBQ4_K_M | — | |
| 2B | 3.3 GBQ4_K_M | — | |
| 4B | 4.5 GBQ4_K_M | — | |
| 26B | 20.0 GBQ4_K_M | — | |
| 31B | 22.0 GBQ4_K_M | — | |
| 2B | 4.0 GBQ4_K_M | — | |
| 4B | 6.0 GBQ4_K_M | — | |
| 744B | 300.0 GBQ2_K | — | |
| 754B | 305.0 GBQ2_K | — | |
| 753B | 274.0 GBQ2_K | — | |
| 117B | 70.0 GBMXFP4 | — | |
| 21B | 15.0 GBMXFP4 | — | |
| 8B | 10.0 GBQ8_0 | — | |
| 20B | 12.0 GBQ4_K_M | — | |
| 7B | 5.5 GBQ4_K_M | — | |
| 1040B | 390.0 GBQ2_K | — | |
| 1000B | 360.0 GBQ2_K | — | |
| 1000B | 359.0 GBQ2_K | — | |
| 33B | 22.0 GBQ4_K_M | — | |
| 405B | 244.5 GBQ4_K_M | — | |
| 70B | 43.5 GBQ4_K_M | — | |
| 8B | 10.0 GBQ8_0 | — | |
| 1B | 3.0 GBQ8_0 | — | |
| 3B | 5.0 GBQ8_0 | — | |
| 11B | 8.5 GBQ4_K_M | — | |
| 90B | 50.0 GBQ4_K_M | — | |
| 70B | 43.5 GBQ4_K_M | — | |
| 400B | 228.0 GBQ4_K_M | — | |
| 109B | 72.0 GBQ4_K_M | — | |
| 24B | 17.0 GBQ4_K_M | — | |
| 428B | 163.0 GBQ2_K | — | |
| 7B | 9.0 GBQ8_0 | — | |
| 123B | 67.0 GBQ4_K_M | — | |
| 12B | 9.5 GBQ4_K_M | — | |
| 24B | 18.0 GBQ4_K_M | — | |
| 141B | 86.0 GBQ4_K_M | — | |
| 47B | 29.7 GBQ4_K_M | — | |
| 8B | 7.5 GBQ4_K_M | — | |
| 253B | 155.0 GBQ4_K_M | — | |
| 34B | 19.0 GBQ4_K_M | — | |
| 8B | 6.0 GBQ4_K_M | — | |
| 7B | 6.9 GBQ4_K_M | — | |
| 3.8B | 5.8 GBQ8_0 | — | |
| 14B | 9.9 GBQ4_K_M | — | |
| 3.8B | 4.5 GBQ4_K_M | — | |
| 14B | 11.0 GBQ4_K_M | — | |
| 14B | 9.9 GBQ4_K_M | — | |
| 32B | 20.7 GBQ4_K_M | — | |
| 72B | 44.7 GBQ4_K_M | — | |
| 7B | 9.0 GBQ8_0 | — | |
| 14B | 12.0 GBQ4_K_M | — | |
| 32B | 23.0 GBQ4_K_M | — | |
| 7B | 9.0 GBQ8_0 | — | |
| 72B | 41.0 GBQ4_K_M | — | |
| 7B | 7.0 GBQ4_K_M | — | |
| 0.6B | 2.5 GBQ4_K_M | — | |
| 14B | 12.0 GBQ4_K_M | — | |
| 235B | 138.0 GBQ4_K_M | — | |
| 30B | 22.0 GBQ4_K_M | — | |
| 32B | 23.0 GBQ4_K_M | — | |
| 4B | 4.5 GBQ4_K_M | — | |
| 8B | 7.5 GBQ4_K_M | — | |
| 0.8B | 1.5 GBQ4_K_M | — | |
| 122B | 85.0 GBQ4_K_M | — | |
| 27B | 19.0 GBQ4_K_M | — | |
| 2B | 3.0 GBQ4_K_M | — | |
| 35B | 12.0 GBQ4_K_M | — | |
| 4B | 4.5 GBQ4_K_M | — | |
| 9B | 7.5 GBQ4_K_M | — | |
| 27B | 20.0 GBQ4_K_M | — | |
| 35B | 27.0 GBQ4_K_M | — | |
| 32B | 21.5 GBQ4_K_M | — | |
| 1.7B | 2.7 GBQ8_0 | — | |
| 15B | 17.0 GBQ8_0 | — | |
| 3B | 3.5 GBQ4_K_M | — | |
| 7B | 5.5 GBQ4_K_M | — | |
| 33B | 22.0 GBQ4_K_M | — | |
| 7B | 6.9 GBQ4_K_M | — | |
| 34B | 21.0 GBQ4_K_M | — | |
| 6B | 5.0 GBQ4_K_M | — | |
| 9B | 6.5 GBQ4_K_M | — | |
| 9B | 8.0 GBQ4_K_M | — |
Methodology: VRAM requirements come from each model's Ollama quantization manifests. Speed estimates are memory-bandwidth-bound (bandwidth ÷ model size) with a penalty when a model spills into CPU offload. Grades weigh speed, VRAM headroom, and model quality.
Just shipped
New & noteworthy
Laguna XS 2.1
70.9 on SWE-bench Verified from 3B active params — frontier-adjacent coding on a 4090.
- 1mo ago
GLM-5.2
The open-weight coding frontier — if you own a 512 GB Mac Studio, this is why.
753Bmin 274 GB - 1mo ago
Kimi K2.7 Code
Moonshot's coding specialist — cheaper agentic runs, but still cluster-or-cloud territory.
1Tmin 359 GB - 1mo ago
MiniMax M3
Frontier coding, 1M context, and video input — and a 192 GB Mac Studio can hold it.
428Bmin 163 GB - 3mo ago
Qwen 3.6 27B
A 27B that out-codes a 397B — the best quality-per-GB on a 24 GB card right now.
27Bmin 20 GB - 3mo ago
Kimi K2.6
A trillion parameters of agentic coding — but for your hardware, it's a cloud tag.
1Tmin 360 GB
Editor's picks
What to run it on
- Budget GPU
NVIDIA GeForce RTX 4060 Ti 16GB
16 GB for the price of 8 — the value entry for mid-size models.
16 GB$499 · 81 models - Mid-range GPU
NVIDIA GeForce RTX 5070 Ti
Fast and roomy enough for most 14–32B models at Q4.
16 GB$749 · 81 models - High-end GPU
NVIDIA GeForce RTX 5090
32 GB of headroom — runs big dense models without flinching.
32 GB$1,999 · 90 models - Budget Mac
Mac mini M4 16GB
The cheapest way into local AI on Apple silicon.
16 GB$599 · 81 models - Mid-range Mac
MacBook Air M5 24GB
Fanless, 24 GB unified — surprisingly capable for its size.
24 GB$1,299 · 84 models - High-end Mac
MacBook Pro M5 Max 128GB
128 GB unified runs models a 5090 can't touch.
128 GB$4,999 · 102 models - Workstation
NVIDIA RTX PRO 6000 Blackwell
96 GB for when consumer cards run out of room.
96 GB$6,800 · 100 models - Value pick
Intel Arc B580
12 GB on a budget — the dark-horse entry card.
12 GB$249 · 63 models
New to local AI?
Install Ollama, pick a model, and start chatting in under ten minutes.