Quantization Size Calculator (GGUF)
Enter a model's size and quant level to estimate its GGUF file size on disk, the VRAM to run it, and the KV-cache for your context length.
Before you download a local model, it helps to know whether it fits your disk and GPU. This gguf size calculator turns two numbers — the parameter count and the quant — into an estimated file size, the VRAM needed to run it, and the KV-cache your context length adds. It answers questions like how big is a Q4 model, how much disk a 70B needs, and what an llm disk size calculator would show across Q2 through FP16. The key detail most quick math misses: Q4_K_M is not 4.0 bits per weight. K-quants store block scales and offsets alongside the weights, so the real figure is about 4.83 bpw and the file is bigger than the name implies. The numbers here are estimates — always check the actual file size on the model card before you commit the download.
Read the guide: GGUF Quantization Sizes ExplainedModel & quant
File size on disk
4.23 GB
7B at Q4_K_M · 4.83 bpw
- Weights
- 4.23 GB
- Estimated VRAM
- 5.08 GB
- KV-cache
- 0 GB
- Bits per weight
- 4.83
| Quant | bpw | Size |
|---|---|---|
| FP16 (full) | 16 | 14 GB |
| Q8_0 | 8.5 | 7.44 GB |
| Q6_K | 6.56 | 5.74 GB |
| Q5_K_M | 5.67 | 4.96 GB |
| Q5_0 | 5.5 | 4.81 GB |
| Q4_K_M | 4.83 | 4.23 GB |
| Q4_0 | 4.5 | 3.94 GB |
| Q3_K_M | 3.91 | 3.42 GB |
| Q2_K | 3.35 | 2.93 GB |
Estimates only. Q4_K_M is about 4.83 bits per weight, not 4.0 — K-quants add block scale and offset overhead, so the file is bigger than the name suggests. VRAM adds a rough 20% for runtime buffers plus the KV-cache, which grows with context length; the KV figure is an order-of-magnitude estimate (~0.5 MB per 1k tokens for a 7B-class model), not architecture-exact. Check the actual GGUF file size on the model card before you download.
How it works
- 1
Enter the parameter count
Type the model size in billions — 7 for a 7B, 13 for a 13B, 70 for a 70B. This is what the weight math scales from.
- 2
Pick the quant level
Choose the GGUF quant, from FP16 down to Q2_K. Each uses its real effective bits per weight, including K-quant overhead, so the size reflects what you would actually download.
- 3
Read the size and compare
The headline is the file size on disk. The stats add weights, estimated VRAM and KV-cache, and the table below sizes the same model across every common quant so you can trade quality for footprint.
Instant & 100% private — nothing is uploaded
Every calculation runs locally in your browser. The prompts, token counts and numbers you enter stay on your own device and are never sent to a server — nothing is stored, logged or shared.
Frequently asked questions
- How big is a Q4_K_M model?
- Roughly the parameter count times 4.83, divided by 8, in gigabytes. A 7B at Q4_K_M is about 4.2 GB; a 13B is about 7.8 GB; a 70B is around 42 GB. That is bigger than a flat 4-bit guess because Q4_K_M actually stores about 4.83 bits per weight.
- Why isn't Q4_K_M exactly 4 bits per weight?
- K-quants group weights into blocks and store a scale and a minimum for each block so they can pack more accuracy into 4-bit weights. That per-block metadata adds overhead, pushing the effective rate to about 4.83 bits per weight. The name refers to the base weight precision, not the whole-file average, which is why real files run larger than the label.
- How much VRAM do I need to run it?
- Enough for the weights plus runtime overhead and the KV-cache. This tool adds a rough 20% on top of the weights for compute buffers and activations, then adds the KV-cache, which grows with your context length. It is a planning estimate — real usage depends on your runtime, batch size and how much you offload to the GPU versus system RAM.
- How exact are these numbers?
- They are estimates. The bits-per-weight figures are realistic averages for each GGUF quant, but exact file sizes vary a little by model architecture and converter, and the KV-cache figure is an order-of-magnitude estimate that depends on layers, heads and cache precision. Use it to judge what fits, then confirm against the real file size listed on the model's page before downloading.
Important
For planning and estimates only. Prices come from a published rate table dated on the page; providers change pricing without notice, and token counts here are approximations. Confirm against the provider’s own pricing before you budget or commit.
More tools
More from the Hivly network
Free sister tools on our other sites.
