Every open model in our directory, ranked by how well it runs on AMD Instinct MI300X (192GB) at 4-bit. Grades weigh whether the model fits in memory and how fast it should run.
AMD Instinct MI300X has 192 GB of VRAM, enough to run 159 of the 169 open models we track at 4-bit. The largest that still fits well is Kimi K2.7 Code (1000B), and the top-rated is Nvidia Nemotron 3 Super at about 41 tok/s.
159 of 169 models run comfortably on AMD Instinct MI300X.
See full AMD Instinct MI300X specs and pricingThe top models this device can run at 4-bit, ranked by fit and speed.
| Model | Grade | Speed | VRAM |
|---|---|---|---|
| Nvidia Nemotron 3 SuperNVIDIA | SS | 41.2 tok/s | 103.5 GB |
| GLM-5Z.ai | SS | 48.6 tok/s | 87.7 GB |
| GLM-5.1Z.ai | SS | 48.6 tok/s | 87.7 GB |
| Kimi K2.7 CodeMoonshot AI | SS | 49.5 tok/s | 86.2 GB |
| Kimi K2.6Moonshot AI |
Nvidia Nemotron 3 Super is the highest-rated model that runs on AMD Instinct MI300X at 4-bit, at about 41 tok/s. In total, AMD Instinct MI300X can run 159 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 |
| 49.5 tok/s |
| 86.2 GB |
| Kimi K2 Instruct 0905Moonshot AI | SS | 50.4 tok/s | 84.6 GB |
| Kimi K2 ThinkingMoonshot AI | SS | 50.4 tok/s | 84.6 GB |
| Kimi K2.5Moonshot AI | SS | 50.4 tok/s | 84.6 GB |
| DeepSeek-V4-FlashDeepSeek | SS | 38.1 tok/s | 112.0 GB |
| Llama 3.1 70B InstructMeta | SS | 37.8 tok/s | 112.8 GB |
| Llama 3.3 70B InstructMeta | SS | 37.8 tok/s | 112.8 GB |
| GLM-4.6Z.ai | SS | 60.7 tok/s | 70.3 GB |
Nvidia Nemotron 3 SuperNVIDIA | 120B(12B active) | SS | 41.2 tok/s | 103.5 GB | |
GLM-5Z.ai | 744B(40B active) | SS | 48.6 tok/s | 87.7 GB | |
GLM-5.1Z.ai | 744B(40B active) | SS | 48.6 tok/s | 87.7 GB | |
Kimi K2.7 CodeMoonshot AI | 1000B(32B active) | SS | 49.5 tok/s | 86.2 GB | |
Kimi K2.6Moonshot AI | 1000B(32B active) | SS | 49.5 tok/s | 86.2 GB | |
Kimi K2 Instruct 0905Moonshot AI | 1000B(32B active) | SS | 50.4 tok/s | 84.6 GB | |
Kimi K2 ThinkingMoonshot AI | 1000B(32B active) | SS | 50.4 tok/s | 84.6 GB | |
Kimi K2.5Moonshot AI | 1000B(32B active) | SS | 50.4 tok/s | 84.6 GB | |
Ad CodeRabbitAI-powered Code Reviews. Cut review time & bugs in half, instantly.Try for Free | |||||
DeepSeek-V4-FlashDeepSeek | 284B(13B active) | SS | 38.1 tok/s | 112.0 GB | |
| 70B | SS | 37.8 tok/s | 112.8 GB | ||
| 70B | SS | 37.8 tok/s | 112.8 GB | ||
GLM-4.6Z.ai | 355B(32B active) | SS | 60.7 tok/s | 70.3 GB | |
Gemma 4 31B ITGoogle | 31B | SS | 52.1 tok/s | 82.0 GB | |
Falcon 180BTechnology Innovation Institute | 180B | SS | 39.6 tok/s | 107.8 GB | |
Mistral Large 3 675BMistral AI | 675B(41B active) | SS | 64.4 tok/s | 66.3 GB | |
Carnice-V2-27bkai-os | 27B | SS | 58.6 tok/s | 72.8 GB | |
Ad PayPerQPay-per-query access to top LLMs without a subscription. Use any model on demand.Try PayPerQ | |||||
Qwen3.6-27BAlibaba | 27B | SS | 58.6 tok/s | 72.8 GB | |
Qwen3.5-27BAlibaba | 27B | SS | 58.6 tok/s | 72.8 GB | |
DeepSeek-V3DeepSeek | 671B(37B active) | SS | 71.3 tok/s | 59.8 GB | |
DeepSeek-R1DeepSeek | 671B(37B active) | SS | 71.3 tok/s | 59.8 GB | |
DeepSeek-V3.1DeepSeek | 671B(37B active) | SS | 71.3 tok/s | 59.8 GB | |
DeepSeek-V3.2DeepSeek | 685B(37B active) | SS | 71.3 tok/s | 59.8 GB | |
Hy3Tencent | 295B(21B active) | SS | 76.6 tok/s | 55.7 GB | |
GLM-4.7Z.ai | 358B(32B active) | SS | 81.1 tok/s | 52.6 GB | |
Ad Vast.aiAffordable on-demand GPU rentals for training and inference. Pick from thousands of hosts.Rent a GPU | |||||
GLM-4.5Z.ai | 355B(32B active) | SS | 82.3 tok/s | 51.8 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.
