Free "fits on X GB" badges
Model creators and repo owners can add a small badge to a Hugging Face or GitHub README showing the smallest common GPU memory size the model fits on. Each badge links back to the model's page here, which has the full numbers and sources.
How the number is worked out. It's the smallest card size in our what can I run tiers (8, 12, 16, 24, 32, 40, 48, 64, 96 or 128 GB) where the model fits comfortably at 4-bit (Q4_K_M) or better with an 8K-token context. It uses the same rule as the rest of the site: real Hugging Face download sizes plus KV cache from the model's own config, with 10% headroom left free. Nothing is estimated by hand. If a model's data changes, the badge updates on the next rebuild.
The badges are free and need no sign-up or tracking. They're plain static SVG files.
| Model | Smallest card size | Copy-paste code |
|---|---|---|
Llama 3.2 3B InstructMeta | 8 GB 8-bit (Q8_0), needs 4.4 GB | |
Llama 3.1 8B InstructMeta | 8 GB 4-bit (Q4_K_M), needs 6.0 GB | |
DeepSeek-R1-0528-Qwen3-8BDeepSeek | 8 GB 4-bit (Q4_K_M), needs 6.2 GB | |
gpt-oss-20bOpenAI | 16 GB 16-bit (FP16/BF16), needs 14.0 GB | |
Qwen3 14BAlibaba Qwen | 12 GB 4-bit (Q4_K_M), needs 10.3 GB | |
Qwen3.8 27BAlibaba Qwen | 24 GB 4-bit (Q4_K_M), needs 18.0 GB | |
Qwen3.8-35B-A3B (Empero Distill)Empero | 32 GB 4-bit (Q4_K_M), needs 21.9 GB | |
Llama 3.1 70B InstructMeta | 64 GB 4-bit (Q4_K_M), needs 45.2 GB | |
gpt-oss-120bOpenAI | 96 GB 16-bit (FP16/BF16), needs 65.7 GB | |
Recursive FlintRecursive | 32 GB 4-bit (Q4_K_M), needs 25.3 GB | |
Ornith-1.5-122BOrnith | 96 GB 4-bit (Q4_K_M), needs 79.9 GB | |
Nanbeige4.3 33BNanbeige | 32 GB 4-bit (Q4_K_M), needs 22.5 GB | |
LFM2.5-24B-A2BLiquid AI | 24 GB 4-bit (Q4_K_M), needs 14.7 GB | |
LFM3 1.7BLiquid AI | 8 GB 16-bit (FP16/BF16), needs 3.6 GB | |
Spark-X2.6 4BXHToken | 8 GB 8-bit (Q8_0), needs 4.7 GB | |
Qwen3.6 35B-A3BAlibaba Qwen | 32 GB 4-bit (Q4_K_M), needs 22.5 GB | |
Mistral 7B Instruct v0.3Mistral AI | 8 GB 4-bit (Q4_K_M), needs 5.4 GB | |
Mixtral 8x7B Instruct v0.1Mistral AI | 32 GB 4-bit (Q4_K_M), needs 27.5 GB | |
Qwen2.5 7B InstructAlibaba Qwen | 8 GB 4-bit (Q4_K_M), needs 5.2 GB | |
Qwen2.5 14B InstructAlibaba Qwen | 12 GB 4-bit (Q4_K_M), needs 10.6 GB | |
Qwen2.5 32B InstructAlibaba Qwen | 32 GB 4-bit (Q4_K_M), needs 22.0 GB | |
Qwen3 32BAlibaba Qwen | 32 GB 4-bit (Q4_K_M), needs 21.9 GB | |
Llama 3.3 70B InstructMeta | 64 GB 4-bit (Q4_K_M), needs 45.2 GB |
Is a badge wrong or a model missing? Use the contact page.