RunsOnMy ModelsGPUs & MacsAPI pricesVRAM calculatorNew

Will this AI model run on my GPU?

VRAM needs for 285 open models on 65 GPUs and Macs, using real GGUF file sizes and each model's own config. Plus live prices for 462 API models, with every price change logged. New models appear within hours of release.

Popular GPUs and Macs

RTX 3060 12GB

12 GB · 171 models fit at 8K context

RTX 4060 Ti 16GB

16 GB · 194 models fit at 8K context

RTX 3090

24 GB · 202 models fit at 8K context

RTX 4090

24 GB · 202 models fit at 8K context

RTX 5090

32 GB · 207 models fit at 8K context

RTX 5070 Ti

16 GB · 194 models fit at 8K context

RX 7900 XTX

24 GB · 202 models fit at 8K context

Mac M4 Pro 48GB

48 GB · 207 models fit at 8K context

Mac M4 Max 128GB

128 GB · 227 models fit at 8K context

Mac M3 Ultra 512GB

512 GB · 278 models fit at 8K context

All 65 GPUs and Macs →

Newest open models

ModelParamsQ4_K_M · 8KOn HF since
SingVL S1 4B inclusionAI4.5B3.6 GBNov 11, 2026
Nex N2.6 nex-agi321B188 GBNov 10, 2026
MiniMax H3 Image PE MiniMaxAI27B16.8 GBNov 8, 2026
Qwen3.8 27B DSpine b16 tencent2.0B2.2 GBNov 7, 2026
Arete 27B NousResearch27B17.4 GBNov 6, 2026
Inkling 2 Small thinkingmachines594B346 GBNov 4, 2026
LongCat 2.5 meituan-longcat1.78T1014 GBNov 4, 2026
S5 nano maritaca-ai4.7B3.7 GBNov 2, 2026
Kimi K3.1 Pro moonshotai2.78T1584 GBNov 2, 2026
Mage VL Cowork microsoft4.7B4.3 GBOct 30, 2026

Everything new this month →

Newest API models

ModelInput / 1MOutput / 1MListed
Tencent: Hy4 flash$0.035$0.141Nov 16, 2026
Qwen: Qwen4 Max$1.425$4.275Nov 16, 2026
inclusionAI: Ling 3.2 Flash (free)freefreeNov 13, 2026
Google: Gemini 4 Hydrogen$1.50$7.50Nov 12, 2026
Google: Gemini 4 Hydrogen (batch)$0.75$3.75Nov 12, 2026
Baidu: ERNIE-5.5-Flash-Preview (free)freefreeNov 10, 2026
Baidu: ERNIE-Light-DT-1.0$1.13$3.94Nov 10, 2026
Google: Gemini 4 Carbon$2.00$10.00Nov 10, 2026

All 462 API prices → · Price change log

Why trust these numbers

Every figure is computed, not guessed: weights from real GGUF file sizes, KV cache from each model's config.json (handles GQA, MLA, sliding-window and hybrid attention, and MoE active vs. total parameters), plus a published overhead. The formulas are on the methodology page, and automated checks hold back any page whose numbers fail a sanity test.