VRAMNeeded Calculator
NEW

Skip the GPU: run open models through Novita AI's model APIs, or give your agents a Novita Sandbox. New categories: AI APIs · Sandboxes → Novita AI is an affiliate partner: we may earn from its links. Disclosure.

Model in, VRAM out.

How much VRAM do you actually need?

Pick a model and your GPU, or enter your GB to see what fits. Every number links to its source.

VRAM calculator

Pick a workload, a model and your graphics card. The answer updates as you choose.

Estimates. LLM numbers use real download sizes from Hugging Face plus the KV cache, the same as our VRAM table and model pages. Checked 2026-10-30.

Too big for your card? Rent the VRAM by the hour, or upgrade.

Run it on a cloud GPU

No upgrade needed

Rent a 24–80 GB GPU for the hour you need it.

Vast.ai Affiliate partner Runpod Affiliate partner E2E Networks Works with

Buy this GPU

Own the VRAM

Compare current GPU prices first on our GPU price table and measured speeds.

B&H Photo Works with

We may earn from links marked Affiliate partner. Disclosure.

New

LLM APIs

Don't want to run it yourself? Compare per-token prices for open models →New

AI agent sandboxes

Isolated sandboxes for agents that run code →

Rent a GPU or buy one

We may earn from some links (disclosure).

Logos are trademarks of their respective owners.

Model pages

DeepSeekDeepSeek-R1-0528-Qwen3-8B8.2B paramsQ4_K_M 5.0 GB128K contextSee what it fits on →DeepSeekDeepSeek-V4.1-Flash-1017763B params1024K contextSee what it fits on →RecursiveRecursive Flint39.8B params256K contextSee what it fits on →GoogleGemma 4 26B A4B25.8B (3.80B active) paramsQ4_K_M 17.0 GB128K contextSee what it fits on →GoogleGemma 4 31B31.3B paramsQ4_K_M 19.6 GB128K contextSee what it fits on →Z.ai (Zhipu AI)GLM-5.3-Flash321B (18B active) params1024K contextSee what it fits on →OpenAIgpt-oss-120b116.8B (5.1B active) paramsQ4_K_M 62.8 GB128K contextSee what it fits on →OpenAIgpt-oss-20b20.9B (3.60B active) paramsQ4_K_M 11.7 GB128K contextSee what it fits on →Moonshot AIKimi K32.8T (104B active) params1024K contextSee what it fits on →Liquid AILFM2.5-24B-A2B24B (2.45B active) paramsQ4_K_M 14.5 GB128K contextSee what it fits on →Liquid AILFM3 1.7B1.7B paramsQ4_K_M 1.1 GB256K contextSee what it fits on →MetaLlama 3.1 70B70.6B paramsQ4_K_M 42.5 GB128K contextSee what it fits on →MetaLlama 3.1 8B8.0B paramsQ4_K_M 4.9 GB128K contextSee what it fits on →MetaLlama 3.2 3B3.2B paramsQ4_K_M 2.0 GB128K contextSee what it fits on →MetaLlama 3.3 70B Instruct70.6B paramsQ4_K_M 42.5 GB128K contextSee what it fits on →Mistral AIMistral 7B Instruct v0.37.2B paramsQ4_K_M 4.4 GB32K contextSee what it fits on →Mistral AIMistral Large 4 (Le Chonk)1.05T (49B active) paramsQ4_K_M 634.8 GB1024K contextSee what it fits on →Mistral AIMixtral 8x7B Instruct v0.146.7B (≈13B active) paramsQ4_K_M 26.4 GB32K contextSee what it fits on →NanbeigeNanbeige4.3 33B33.1B paramsQ4_K_M 20.8 GB512K contextSee what it fits on →NVIDIANemotron 3.5 Transcribe 30B-A4B34B (4.3B active) params64K contextSee what it fits on →OrnithOrnith-1.5-122B125B params256K contextSee what it fits on →Alibaba QwenQwen2.5 14B Instruct14.8B paramsQ4_K_M 9.0 GB128K contextSee what it fits on →Alibaba QwenQwen2.5 32B Instruct32.8B paramsQ4_K_M 19.9 GB128K contextSee what it fits on →Alibaba QwenQwen2.5 7B Instruct7.6B paramsQ4_K_M 4.7 GB128K contextSee what it fits on →Alibaba QwenQwen3 14B14.8B paramsQ4_K_M 9.0 GB128K contextSee what it fits on →Alibaba QwenQwen3 32B32.8B paramsQ4_K_M 19.8 GB128K contextSee what it fits on →Alibaba QwenQwen3.6 35B-A3B36.0B (≈3B active) paramsQ4_K_M 22.3 GB256K contextSee what it fits on →Alibaba QwenQwen3.8 27B27.8B paramsQ4_K_M 17.4 GB256K contextSee what it fits on →Empero (community distill of Alibaba Qwen)Qwen3.8-35B-A3B (Empero Distill)35.1B (≈3B active) paramsQ4_K_M 21.7 GB256K contextSee what it fits on →Alibaba Qwen (pruned by ISTA-DASLab / AgentionAI)Qwen3.8-Flash-Next Coder176.9B base, half the experts kept (6B active in the base) params256K contextSee what it fits on →Alibaba QwenQwen3.8-Flash-Next180B (125B MoE + 51B n-gram + 4B MTP; 6B active) paramsQ4_K_M 119.6 GB256K contextSee what it fits on →XHTokenSpark-X2.6 4B4.1B paramsQ4_K_M 3.5 GB1024K contextSee what it fits on →

Start here

What can I run?Pick a GPU (or VRAM size) and see which models fit. Compare modelsSide-by-side VRAM for common local models. Gaming VRAMMemory needs by resolution — 1080p, 1440p, 4K. GPU VRAM lookupEvery GPU and Mac: memory, which LLMs fit, rental price. GPU pricesWhat cards cost to buy or rent by the hour. VRAM by precisionFP16, FP8, INT8 and INT4 memory for any model or parameter count.

How every page works

Quick answer firstOne paragraph with the number you came for.
Real file sizesTaken from the actual downloads on Hugging Face, not guessed from parameter counts.
Context countsKV cache from the model own config — a long context can double the memory you need.
Sources + checked dateOn every page. If we cannot source a number, we leave it out.