Can Your PC Run a Local AI Model?
Pick a model, quantization, context length and graphics card to see if it fits in memory.
Selection models released 2 Jun 2026 to 2 Oct 2026 (dated cohort) Newest model carried 22 Sep 2026 Baseline sources checked 15 Sep 2026 Latest source check 3 Nov 2026
Models: recent releases, plus the most downloaded on Hugging Face. Last verified 3 Nov 2026. Check dates are in the tables below.
Covers releases up to 2 Oct 2026. Newer ones are added after a source check.
CHECK A CONFIGURATION
16 MODELS / 5 PROFILESGemma 4 12B on 8 GiB graphics card
Estimated memory against usable target
7.2 GiB usable of 8 GiBGoogle describes this as a local multimodal model and publishes downloadable checkpoints.
Estimate only: memory fit is not a speed, output-quality, runtime-support, or compatibility guarantee.
Memory needed = weights + context + overhead
How it's calculated
- Weights: parameters × bits ÷ 8, plus 10%.
- Context: grows with the context length you pick.
- Overhead: a set amount per model for the runtime.
- Reserve: 10% of your memory stays free.
Based on Ollama's GPU docs and llama.cpp's quantization notes.
Not a guarantee
- A pass means it should fit in memory.
- It may still fail to load.
- It says nothing about speed.
- Long contexts and some runtimes need more.
Before you download
- Set the quantization and context you will really use.
- Check the model card.
- Check your runtime supports the format.
Q4-Q8 and FP16 are generic sizes. Some models ship their own formats (for example NVFP4, MXFP4, INT4 or FP8), so their estimate is approximate.
Hosted models instead? Compare provider plans. Which runtimes are active? Open-source AI tracker.
REGISTER SOURCES
RELEASES 2 Jun 2026 TO 2 Oct 2026 / MOST DOWNLOADED| Model | Released | Weights | Checked |
|---|---|---|---|
| Gemma 4 12B | 3 Jun 2026 | 2 Nov 2026 | |
| NVIDIA Nemotron 3 Ultra | 4 Jun 2026 | 2 Nov 2026 | |
| NVIDIA Nemotron 3.5 Lightning 30B-A3B | 11 Aug 2026 | NVFP4 | 2 Nov 2026 |
| NVIDIA Nemotron 3.5 Content Safety | 2 Jun 2026 | BF16 | 2 Nov 2026 |
| NVIDIA Nemotron 3.5 ASR Streaming 0.6B | 4 Jun 2026 | BF16 | 2 Nov 2026 |
| DeepSeek-V4-Flash-0731 | 31 Jul 2026 | FP8 (FP4 cache options) | 2 Nov 2026 |
| Kimi K3 | 16 Jul 2026 | MXFP4 weights / MXFP8 activations (QAT) | 15 Sep 2026 |
| Kimi K2.7 Code | 16 Jul 2026 | Native INT4 | 2 Nov 2026 |
| Qwen3 12B, E-AI compressed V1.5 | 1 Jul 2026 | 2 Nov 2026 | |
| Qwen3.8-27B release note | 14 Aug 2026 | BF16 | 2 Nov 2026 |
| MiMo-V2.6-Distill-Qwen-9B release note | 22 Sep 2026 | BF16 | 2 Nov 2026 |
| Model | Released | Weights | Checked |
|---|---|---|---|
| Qwen3-14B release note | 29 Apr 2025 | BF16 | 3 Nov 2026 |
| Gemma 4 31B IT release note | 2 Apr 2026 | BF16 | 3 Nov 2026 |
| gpt-oss-20b release note | 5 Aug 2025 | MXFP4 (MoE weights), BF16 other layers | 3 Nov 2026 |
| Llama 3.1 8B Instruct release note | 23 Jul 2024 | BF16 | 3 Nov 2026 |
| NVIDIA Nemotron 3 Nano 4B release note | 16 Mar 2026 | BF16 | 3 Nov 2026 |
| Memory | Example cards | Checked |
|---|---|---|
| 8 GiB VRAM | GeForce RTX 4060 | 2 Nov 2026 |
| 12 GiB VRAM | GeForce RTX 4070 SUPER | 2 Nov 2026 |
| 16 GiB VRAM | GeForce RTX 4080 SUPER Radeon RX 9070 XT GeForce RTX 4060 Ti 16GB |
2 Nov 2026 |
| 24 GiB VRAM | GeForce RTX 4090 Radeon RX 7900 XTX |
2 Nov 2026 |
| 32 GiB VRAM | GeForce RTX 5090 | 2 Nov 2026 |