What can you run on an 8GB graphics card?
With 8GB VRAM you can run 57 of the 129 open-source models we track — up to about 16B parameters at Q4 quantization. The most popular that fits is Qwen3.5-9B, needing 6.8 GB and generating around ~31 tok/s on typical hardware of this size.
Typical machines: RTX 4060 · RTX 3070 · RTX 3060 Ti · RX 7600. After the operating system takes its share, about 7.7 GB is free for a model.
Having the memory is not the same as having the speed
8GB VRAM decides whether a model loads. How fast it answers is decided by memory bandwidth — how quickly your machine can read the model, which it must do once for every word it writes. Machines with 8GB VRAM span 272–448 GB/s, so the fastest is about 1.6× quicker than the slowest with identical capacity. The table quotes a RTX 4060; pick yours below to see the difference. Why bandwidth and not the processor →
Models that run on 8GB VRAM in 2026
| Model↕ | Size↕ | Memory↑ | Est. speed↕ | Downloads↕ | Feels like |
|---|---|---|---|---|---|
| granite-embedding-97m-multilingual-r2Ibm | 97M | 603 MBRuns well | ~3005 tok/s | 97,079 | Fast |
| LFM2.5-Encoder-230MLiquid Ai | 230M | 793 MBRuns well | ~1275 tok/s | 13,432 | Fast |
| LFM2.5-230MLiquid Ai | 230M | 808 MBRuns well | ~1152 tok/s | 55,999 | Fast |
| granite-embedding-311m-multilingual-r2Ibm | 312M | 821 MBRuns well | ~698 tok/s | 64,530 | Fast |
| granite-speech-5.0-470m-turboctcIbm | 473M | 857 MBRuns well | ~619 tok/s | 11,294 | Fast |
| harrier-oss-v1-270mMicrosoft | 268M | 865 MBRuns well | ~698 tok/s | 727,222 | Fast |
| LFM2.5-Encoder-350M-Policy-LinterLiquid Ai | 355M | 885 MBRuns well | ~825 tok/s | 9,957 | Fast |
| LFM2.5-VL-450MLiquid Ai | 449M | 900 MBRuns well | ~771 tok/s | 67,396 | Fast |
| nemotron-3.5-asr-streaming-0.6bNVIDIA | 638M | 1.1 GBRuns well | ~357 tok/s | 917,259 | Fast |
| Qwen3.5-0.8BAlibaba / Qwen | 873M | 1.3 GBRuns well | ~332 tok/s | 2,881,964 | Fast |
| privacy-filterOpenAI · MoE | 1.4B | 1.4 GBRuns well | ~1046 tok/s280M active | 456,933 | Fast |
| LFM2.5-1.2B-InstructLiquid Ai | 1.2B | 1.4 GBRuns well | ~242 tok/s | 366,847 | Fast |
| LFM2.5-VL-1.6BLiquid Ai | 1.6B | 1.4 GBRuns well | ~242 tok/s | 248,110 | Fast |
| Nemotron-3-Embed-1B-BF16NVIDIA | 1.1B | 1.5 GBRuns well | ~257 tok/s | 386,914 | Fast |
| Hy-MT2-1.8BTencent | 2B | 1.9 GBRuns well | ~156 tok/s | 24,475 | Fast |
| MiniMax-Music3MiniMax | 2.4B | 2.1 GBRuns well | ~127 tok/s | 23,217 | Fast |
| Qwen3.5-2BAlibaba / Qwen | 2.3B | 2.1 GBRuns well | ~127 tok/s | 3,097,484 | Fast |
| cohere-transcribe-arabic-07-2026Cohere | 2.1B | 2.3 GBRuns well | ~113 tok/s | 51,893 | Fast |
| cohere-transcribe-03-2026Cohere | 2.1B | 2.3 GBRuns well | ~113 tok/s | 607,964 | Fast |
| granite-speech-4.1-2bIbm | 2.3B | 2.4 GBRuns well | ~113 tok/s | 237,181 | Fast |
| LFM2.5-2.6BLiquid Ai | 2.7B | 2.5 GBRuns well | ~106 tok/s | 133,677 | Fast |
| LFM2.5-VL-3BLiquid Ai | 3.1B | 2.5 GBRuns well | ~106 tok/s | 24,282 | Fast |
| granite-4.0-1b-speechIbm | 2.3B | 2.5 GBRuns well | ~110 tok/s | 74,644 | Fast |
| LFM2.5-Audio-1.5BLiquid Ai | 1.5B | 2.6 GBRuns well | ~97 tok/s | 1,702 | Fast |
| Unlimited-OCRBaidu AI | 3.3B | 2.7 GBRuns well | ~91 tok/s | 3,008,635 | Fast |
| Nemotron-Labs-Diffusion-3BNVIDIA | 3B | 2.8 GBRuns well | ~98 tok/s | 35,414 | Fast |
| LocateAnything-3BNVIDIA | 3.8B | 2.8 GBRuns well | ~84 tok/s | 95,054 | Fast |
| granite-4.1-3bIbm | 3.4B | 2.9 GBRuns well | ~86 tok/s | 122,076 | Fast |
| granite-4.2-3bIbm | 3.7B | 3.1 GBRuns well | ~79 tok/s | 14,073 | Fast |
| granite-4.0-3b-visionIbm | 4B | 3.1 GBRuns well | ~79 tok/s | 1,069 | Fast |
| Shieldstral-1.0-3BMistral AI | 3.8B | 3.1 GBRuns well | ~82 tok/s | 20,895 | Fast |
| Ministral-3-3B-Instruct-2512Mistral AI | 3.8B | 3.1 GBRuns well | ~82 tok/s | 544,419 | Fast |
| Nemotron-Labs-Diffusion-3B-BaseNVIDIA | 3.8B | 3.3 GBRuns well | ~76 tok/s | 19,213 | Fast |
| Cosmos3-EdgeNVIDIA | 3.9B | 3.3 GBRuns well | ~76 tok/s | 1,325,763 | Fast |
| Nemotron-3.5-Content-SafetyNVIDIA | 4.3B | 3.6 GBRuns well | ~71 tok/s | 134,547 | Fast |
| gemma-4-E2B-itGoogle DeepMind | 5.1B | 3.8 GBRuns well | ~57 tok/s | 3,235,641 | Fast |
| Qwen3.5-4BAlibaba / Qwen | 4.7B | 3.8 GBRuns well | ~65 tok/s | 7,681,585 | Fast |
| Fara1.5-4BMicrosoft | 4.5B | 4.0 GBRuns well | ~61 tok/s | 2,786 | Fast |
| MolmoWeb-4BAi2 | 4.9B | 4.1 GBRuns well | ~60 tok/s | 2,152 | Fast |
| NVIDIA-Nemotron-3-Nano-4B-BF16NVIDIA | 4B | 4.1 GBRuns well | ~62 tok/s | 712,701 | Fast |
| Fara-7BMicrosoft | 8.3B | 5.5 GBRuns well | ~38 tok/s | 1,593 | Fast |
| Hy-MT2-7BTencent | 8B | 5.7 GBRuns well | ~38 tok/s | 12,004 | Fast |
| LFM2.5-8B-A1BLiquid Ai · MoE | 8.5B | 5.9 GBRuns well | ~290 tok/s1B active | 96,675 | Fast |
| Nemotron-3-Embed-8B-BF16NVIDIA | 8B | 5.9 GBRuns well | ~37 tok/s | 81,804 | Fast |
| Nemotron-Labs-Diffusion-8BNVIDIA | 8B | 5.9 GBRuns well | ~37 tok/s | 163,233 | Fast |
| Cosmos3-NanoNVIDIA | 16B | 6.2 GBRuns well | ~35 tok/s | 244,209 | Fast |
| gemma-4-E4B-itGoogle DeepMind | 8B | 6.2 GBRuns well | ~33 tok/s | 4,740,694 | Fast |
| Nemotron-Labs-Diffusion-8B-BaseNVIDIA | 8.5B | 6.2 GBRuns well | ~34 tok/s | 280,417 | Fast |
| Ministral-3-8B-Instruct-2512Mistral AI | 8.9B | 6.3 GBRuns well | ~34 tok/s | 119,379 | Fast |
| granite-4.1-8bIbm | 8.8B | 6.3 GBRuns well | ~35 tok/s | 1,441,995 | Fast |
| MolmoWeb-8BAi2 | 8.7B | 6.4 GBRuns well | ~34 tok/s | 1,679 | Fast |
| granite-4.2-8bIbm | 8.8B | 6.6 GBRuns well | ~33 tok/s | 12,050 | Fast |
| Ornith-1.0-9BDeepreinforce | 9B | 6.7 GBTight | ~31 tok/s | 2,250,422 | Fast |
| Qwen3.5-9BAlibaba / Qwen | 9.7B | 6.8 GBTight | ~31 tok/s | 13,660,443 | Fast |
| GLM-4.6V-FlashZhipu AI | 10B | 7.0 GBTight | ~29 tok/s | 282,562 | Fast |
| Fara1.5-9BMicrosoft | 9.4B | 7.0 GBTight | ~30 tok/s | 4,461 | Fast |
| Olmo-3-7B-InstructAi2 | 7.3B | 7.2 GBTight | ~40 tok/s | 406,776 | Fast |
Sorted by memory, smallest first — click any column to change it.
Step up to 12GB VRAM and 3 more models come within reach
Including gemma-4-12B-it (12B), Ministral-3-14B-Instruct-2512 (14B) and Nemotron-Labs-Diffusion-14B (14B) — around ~45 tok/s on a typical machine that size.
See everything 12GB VRAM runs →All figures at Q4_K_M quantization, 4K context. “Memory” includes the model, its context cache and runtime overhead. Speed is calculated, not benchmarked: 272 GB/s × 0.65 ÷ bytes read per token, assuming a RTX 4060. Scale it by your own machine's bandwidth from the list above. Mixture-of-experts models read only their active parameters per token, which is why some large ones outrun smaller dense models.
Image generators that also fit
These make pictures rather than text. Their memory use scales with the resolution you generate at, not conversation length, so treat these figures as a floor rather than a ceiling.
What you'd need more memory for
The next models up, and what they ask for.