NVIDIA · Blackwell · 24 GB VRAM tier
Best AI Models for NVIDIA RTX PRO 4000 Blackwell (24 GB)
Runs 27–31B and 35B MoE models well; 70B doesn't fit in 24 GB.
| # | Model | Speed | llmrun Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 25.1 t/s | 25.1 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 25.1 t/s | 25.1 tok/s | Q4_K_M · 17.9 GB | |
| 3 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 21.4 t/s | 21.4 tok/s | Q4_K_M · 21.7 GB | |
| 4 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 126 t/s | 126 tok/s | Q4_K_M · 16.5 GB | |
| 5 | Q4_K_M · 19.3 GB · 162 t/s | 162 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Coding Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 25.1 t/s | 25.1 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 21.4 t/s | 21.4 tok/s | Q4_K_M · 21.7 GB | |
| 3 | Qwen3.6 35B A3BAlibaba Q4_K_M · 22.0 GB · 113 t/s | 113 tok/s | Q4_K_M · 22.0 GB | |
| 4 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 126 t/s | 126 tok/s | Q4_K_M · 16.5 GB | |
| 5 | Q4_K_M · 19.3 GB · 162 t/s | 162 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Reasoning Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 25.1 t/s | 25.1 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 21.4 t/s | 21.4 tok/s | Q4_K_M · 21.7 GB | |
| 3 | Qwen3.5 27BAlibaba Q4_K_M · 17.9 GB · 25.1 t/s | 25.1 tok/s | Q4_K_M · 17.9 GB | |
| 4 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 126 t/s | 126 tok/s | Q4_K_M · 16.5 GB | |
| 5 | Q4_K_M · 19.3 GB · 162 t/s | 162 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Math Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 25.1 t/s | 25.1 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 25.1 t/s | 25.1 tok/s | Q4_K_M · 17.9 GB | |
| 3 | Qwen3.6 35B A3BAlibaba Q4_K_M · 22.0 GB · 113 t/s | 113 tok/s | Q4_K_M · 22.0 GB | |
| 4 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 21.4 t/s | 21.4 tok/s | overall | Q4_K_M · 21.7 GB |
| 5 | Q4_K_M · 19.3 GB · 162 t/s | 162 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Long context Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 25.1 t/s | 25.1 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 25.1 t/s | 25.1 tok/s | Q4_K_M · 17.9 GB | |
| 3 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 126 t/s | 126 tok/s | Q4_K_M · 16.5 GB | |
| 4 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 138 t/s | 138 tok/s | Q4_K_M · 13.1 GB | |
| 5 | LFM2.5 24B A2BLiquid AI Q4_K_M · 15.0 GB · 176 t/s | 176 tok/s | Q4_K_M · 15.0 GB |
| # | Model | Speed ↓ | llmrun Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | GPT OSS 20BOpenAIHow to run → Q4_K_M · 13.1 GB · 138 t/s | 138 tok/s | Q4_K_M · 13.1 GB | |
| 2 | Q4_K_M · 19.3 GB · 162 t/s | 162 tok/s | Q4_K_M · 19.3 GB | |
| 3 | Qwen3 30B A3B Thinking 2507Alibaba Q4_K_M · 19.4 GB · 136 t/s | 136 tok/s | Q4_K_M · 19.4 GB | |
| 4 | Qwen3 30B A3B Instruct 2507Alibaba Q4_K_M · 19.4 GB · 136 t/s | 136 tok/s | Q4_K_M · 19.4 GB | |
| 5 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 126 t/s | 126 tok/s | Q4_K_M · 16.5 GB |
| # | Model | Speed | Vision Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 25.1 t/s | 25.1 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 25.1 t/s | 25.1 tok/s | Q4_K_M · 17.9 GB | |
| 3 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 21.4 t/s | 21.4 tok/s | Q4_K_M · 21.7 GB | |
| 4 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 126 t/s | 126 tok/s | Q4_K_M · 16.5 GB | |
| 5 | JEV 27B VLautotrust Q4_K_M · 17.9 GB · 25.1 t/s | 25.1 tok/s | Q4_K_M · 17.9 GB |
The RTX PRO 4000 Blackwell runs 27–31B and 35B MoE models well. Blackwell generation, released 2025.
RTX PRO 4000 Blackwell strengths and weaknesses
Strengths
- Runs 35B MoE models: Qwen3.6 35B A3B at 113 tok/s in 22 GB
- Room for long context: Qwen3.8 27B at Q4 uses 17.9 GB, leaving 6.1 GB free
- Low power draw: 145 W board power; a 300 W PSU is enough
Weaknesses
- 70B models don't fit: Llama 3.3 70B Instruct needs 51.3 GB at Q4 vs 24 GB here
Output speed depends on memory bandwidth (672 GB/s). Grey rows need more than 24 GB.
- 9BQwen3.5 9B · Q4_K_M · 6.6 GB68.7tok/s
- 14BQwen3 14B · Q4_K_M · 11.9 GB46.0tok/s
- 27–31BQwen3.8 27B · Q4_K_M · 17.9 GB25.1tok/s
- 35B MoEQwen3.6 35B A3B · Q4_K_M · 22.0 GB113tok/s
- 70BLlama 3.3 70B Instruct · not at 4-bitneeds 51.3 GB
- 120B MoEGPT OSS 120B · not at 4-bitneeds 70.8 GB
Estimated speeds: Q4_K_M, 4K context, one user. How we estimate
All 1141 models on the RTX PRO 4000 Blackwell
Featured onlyBest quant that fits, ordered by llmrun Score. Fit grades how well it runs here, S best. How ranking works
| Model | Quant | VRAM | Speed | Context | Score | Fit |
|---|---|---|---|---|---|---|
Kimi K2.6 Eagle3Chat | Q4_K_M | 1.9 GB | 299.2 t/s | 262K | S100, Runs great | |
| Q4_K_M | 8.9 GB | 53.9 t/s | 131K | A81, Runs well | ||
| Q4_K_M | 13.1 GB | 137.9 t/s | 131K | S96, Runs great | ||
QwQ 32BChatReasoningFunctions | Q4_K_S | 23.0 GB | 22.7 t/s | 41K | C38, Usable | |
Carbon 8BChat | Q4_K_M | 7.4 GB | 79.1 t/s | 33K | S87, Runs great | |
Supra Mini V5 8MChat | Q4_K_M | 0.3 GB | 1365.0 t/s | 1K | S100, Runs great | |
Soren 1 SmallChatReasoningCodeMath | BF16 | 4.3 GB | 105.0 t/s | 1049K | S92, Runs great | |
| BF16 | 10.0 GB | 49.8 t/s | 131K | A79, Runs well | ||
AI21 Jamba Reasoning 3BChatReasoningFunctions | Q4_K_M | 2.3 GB | 194.1 t/s | 262K | S99, Runs great | |
| Q4_K_M | 0.6 GB | 873.6 t/s | 8K | S100, Runs great | ||
Tiny Aya GlobalChat | BF16 | 9.4 GB | 59.3 t/s | — | A84, Runs well | |
| Q4_K_M | 9.2 GB | 52.2 t/s | 131K | A81, Runs well | ||
GPT X2 125MChat | BF16 | 0.6 GB | 682.5 t/s | 1K | S100, Runs great | |
Deepseek Vl2 TinyVision | BF16 | 9.4 GB | 116.8 t/s | — | S94, Runs great | |
Vllm Translategemma 12B ITChatTranslation | Q4_K_M | 9.8 GB | 48.7 t/s | 131K | A79, Runs well |
Rent NVIDIA RTX PRO 4000 Blackwell in the cloud
Hourly rate on demand for one RTX PRO 4000 Blackwell.
Vast.aiMarketplace
from$0.26/hr
Buying one ($1,500) costs as much as ~5,800 hours of rental, about 24 months at 8 hours a day.
Cost per million tokens
Output-token cost on an RTX PRO 4000 Blackwell rented at $0.26/hr, one request at a time. Batch several users on the same card and the cost per token drops several-fold.
| Model | Quant | Speed | Per 1M output |
|---|---|---|---|
| NVIDIA Nemotron 3.5 Lightning 30B A3B BF16 | 162 tok/s | $0.44$0.0051 input | |
| Gemma 4 26B A4B IT | 126 tok/s | $0.57$0.0095 input | |
| GLM 4.7 Flash | 108 tok/s | $0.67$0.010 input | |
| Qwen3.8 27B | 25 tok/s | $2.86$0.060 input | |
| Muse Glimmer 30B | 24 tok/s | $3.02$0.065 input |
Specs
- Architecture
- Blackwell
- VRAM
- 24 GB GDDR7
- Memory bandwidth
- 672.0 GB/s
- FP16 (dense)
- 147.25 TFLOPS
- CUDA cores
- 8,960
- Tensor cores
- 280
- Compute capability
- 12.0
- TDP
- 145 W
- Released
- 2025-04-01
- MSRP
- $1,500
RTX PRO 4000 Blackwell FAQ
What is the largest model the RTX PRO 4000 Blackwell can run?
Qwen3.6 35B A3B (36.0B) is the largest at 4-bit or better: at Q4_K_M it uses 22.0 GB of 24 GB, and runs at about 113 tok/s.
Can the RTX PRO 4000 Blackwell run Qwen3.8 27B?
Yes. Qwen3.8 27B fits at Q4_K_M, using 17.9 GB of 24 GB, and it is the best model that runs well on this card.
Can the RTX PRO 4000 Blackwell run 70B models?
No. Llama 3.3 70B Instruct needs 51.3 GB at Q4, more than the 24 GB on this card.
How many tokens per second does the RTX PRO 4000 Blackwell give?
About 25.1 tok/s on Qwen3.8 27B at Q4_K_M, with the first token after 3.4 s on a 4K-token prompt.
What power supply does the RTX PRO 4000 Blackwell need?
Plan a 300 W PSU (2× the 145 W TDP).
Is 24 GB enough for coding and agents?
Yes for coding: Qwen3.8 27B, the best of the 9 coding models that fit, runs at about 25 tok/s. For agents, not yet: no agent-scored model fits in less than 96 GB.