Every open model in our directory, ranked by how well it runs on NVIDIA GeForce RTX 4060 (8GB) at 4-bit. Grades weigh whether the model fits in memory and how fast it should run.
NVIDIA GeForce RTX 4060 has 8.0 GB of VRAM, enough to run 87 of the 169 open models we track at 4-bit. The largest that still fits well is Qwen3-30B-A3B (30B), and the top-rated is Qwen3-30B-A3B at about 41 tok/s.
87 of 169 models run comfortably on NVIDIA GeForce RTX 4060.
See full NVIDIA GeForce RTX 4060 specs and pricingThe top models this device can run at 4-bit, ranked by fit and speed.
| Model | Grade | Speed | VRAM |
|---|---|---|---|
| Qwen3-30B-A3BAlibaba | SS | 40.7 tok/s | 5.4 GB |
| Llama 3 8B InstructMeta | SS | 38.7 tok/s | 5.7 GB |
| Carnice-9b for Hermes agentkai-os | SS | 36.4 tok/s | 6.0 GB |
| LFM2.5-8B-A1BLiquid AI | SS | 75.3 tok/s | 2.9 GB |
| PersonaPlex 7B |
Qwen3-30B-A3B is the highest-rated model that runs on NVIDIA GeForce RTX 4060 at 4-bit, at about 41 tok/s. In total, NVIDIA GeForce RTX 4060 can run 87 of the 169 open models we track.
We size each model at 4-bit (Q4_K_M), add the KV cache for its context length and a fixed runtime overhead, then compare that to this device VRAM. Each model is graded on how comfortably it fits and how fast it should run on this card.
It is an estimate of decode speed in tokens per second at 4-bit, based on this device memory bandwidth and each model size. Real speed depends on your runtime, batch size, and context length, so use it to compare models, not as a hard promise.
Lower-bit quantization shrinks a model so more of them fit. These grades already assume 4-bit, which is the common local default. Going below 4-bit can squeeze a larger model in at some quality cost; going above needs more VRAM and may not fit.
| SS |
| 45.7 tok/s |
| 4.8 GB |
| Llama 2 7B ChatMeta | SS | 45.7 tok/s | 4.8 GB |
| VibeThinker-3BWeiboAI | AA | 57.4 tok/s | 3.8 GB |
| Gemma 4 E2B ITGoogle | AA | 59.1 tok/s | 3.7 GB |
| Mistral 7B InstructMistral AI | AA | 34.2 tok/s | 6.4 GB |
| Gemma 4 E4B ITGoogle | AA | 31.7 tok/s | 6.9 GB |
| Gemma 3 4B ITGoogle | AA | 31.7 tok/s | 6.9 GB |
| North Mini CodeCohere | CC | 26.1 tok/s | 8.4 GB |
Qwen3-30B-A3BAlibaba | 30B(3B active) | SS | 40.7 tok/s | 5.4 GB | |
| 8B | SS | 38.7 tok/s | 5.7 GB | ||
| 9B | SS | 36.4 tok/s | 6.0 GB | ||
LFM2.5-8B-A1BLiquid AI | 8.3B(1.5B active) | SS | 75.3 tok/s | 2.9 GB | |
PersonaPlex 7BNVIDIA | 7B | SS | 45.7 tok/s | 4.8 GB | |
Llama 2 7B ChatMeta | 7B | SS | 45.7 tok/s | 4.8 GB | |
VibeThinker-3BWeiboAI | 3B | AA | 57.4 tok/s | 3.8 GB | |
Gemma 4 E2B ITGoogle | 2B | AA | 59.1 tok/s | 3.7 GB | |
Ad CodeRabbitAI-powered Code Reviews. Cut review time & bugs in half, instantly.Try for Free | |||||
Mistral 7B InstructMistral AI | 7B | AA | 34.2 tok/s | 6.4 GB | |
Gemma 4 E4B ITGoogle | 4B | AA | 31.7 tok/s | 6.9 GB | |
Gemma 3 4B ITGoogle | 4B | AA | 31.7 tok/s | 6.9 GB | |
North Mini CodeCohere | 30B(3B active) | CC | 26.1 tok/s | 8.4 GB | |
Nemotron 3 Nano OmniNVIDIA | 30B(3B active) | CC | 25.7 tok/s | 8.5 GB | |
Qwen3.6 35B-A3BAlibaba | 35B(3B active) | CC | 25.7 tok/s | 8.5 GB | |
Qwen3.5-35B-A3BAlibaba | 35B(3B active) | CC | 25.7 tok/s | 8.5 GB | |
Llama 2 13B ChatMeta | 13B | CC | 25.9 tok/s | 8.5 GB | |
Ad PayPerQPay-per-query access to top LLMs without a subscription. Use any model on demand.Try PayPerQ | |||||
| 8B | FF | 16.4 tok/s | 13.3 GB | ||
Qwen3.5-9BAlibaba | 9B | FF | 8.9 tok/s | 24.6 GB | |
Gemma 4 12BGoogle | 12B | FF | 6.8 tok/s | 32.0 GB | |
Gemma 4 12B Coderyuxinlu1 | 12B | FF | 6.8 tok/s | 32.0 GB | |
Mistral Small 3 24BMistral AI | 24B | FF | 5.6 tok/s | 39.0 GB | |
DiffusionGemma 26B-A4BGoogle | 25.2B(3.8B active) | FF | 20.9 tok/s | 10.5 GB | |
Gemma 4 26B-A4B ITGoogle | 26B(4B active) | FF | 19.9 tok/s | 11.0 GB | |
Carnice-V2-27bkai-os | 27B | FF | 3.0 tok/s | 72.8 GB | |
Ad Vast.aiAffordable on-demand GPU rentals for training and inference. Pick from thousands of hosts.Rent a GPU | |||||
Qwen3.6-27BAlibaba | 27B | FF | 3.0 tok/s | 72.8 GB | |
Model Compatibility
Use the calculator to weigh this device against any other, model by model, with speed and fit scores.
Mixture-of-experts models only activate a fraction of their parameters at a time, so their runtime memory is closer to the active size than the headline size. We score fit on the active parameters, which is why some very large models still fit.
