Model reference · open weights
Qwen3 is an open-weight language model from PrimeIntellect. Qwen3-8B (BF16) weighs 16.4 GB; the smallest configuration that runs it is 2× RTX 3060 12 GB.
What it is
| Released by | PrimeIntellect |
|---|---|
| Released | 2025-08-12 |
| Parameters | 752M |
| VRAM | 16.4 GB for the weights |
What it runs on
| Card | Requests at once | Context max | Memory | |
|---|---|---|---|---|
| 8K each | 32K each | |||
| RTX 3060 12 GB … RTX 4060 Ti 16 GB 2 smaller cards | — | — | — | |
| RTX 3090 24 GB | 5 | 1 | all 40K | 23.4 GB |
| RTX 4090 24 GB | 5 | 1 | all 40K | 23.4 GB |
| RTX 5090 32 GB | 11 | 2 | all 40K | 31.0 GB |
| L40S 48 GB | 22 | 5 | all 40K | 44.0 GB |
| A100 80 GB | 50 | 12 | all 40K | 78.2 GB |
| H100 80 GB | 46 | 11 | all 40K | 78.1 GB |
| RTX PRO 6000 Blackwell 96 GB | 59 | 14 | all 40K | 93.8 GB |
| DGX Spark (GB10) 128 GB unified | 71 | 17 | all 40K | 107 GB |
| H200 141 GB | 96 | 24 | all 40K | 138 GB |
| B200 180 GB | 127 | 31 | all 40K | 176 GB |
| 2× RTX 3060 12 GB tensor parallel | 4 | 1 | 35K | 11.6 GB a card |
| 2× RTX 4060 Ti 16 GB tensor parallel | 10 | 2 | all 40K | 15.4 GB a card |
| 2× RTX 4090 24 GB tensor parallel | 23 | 5 | all 40K | 23.4 GB a card |
| 2× RTX 3090 24 GB tensor parallel | 23 | 5 | all 40K | 23.4 GB a card |
| Requests at once | 8K tokens each | 32K tokens each |
|---|---|---|
| 1 | 18.3 GB | 22.0 GB |
| 5 | 23.2 GB | 41.3 GB |
| 8 | 26.8 GB | 55.8 GB |
| 16 | 36.5 GB | 94.4 GB |
| 32 | 55.8 GB | 172 GB |
| 64 | 94.4 GB | 326 GB |
One card, with vLLM's small-card settings.
Builds
| Build | Params | Precision | Weights | Smallest setup |
|---|---|---|---|---|
| Qwen3-0.6B ↗ | 752M | BF16 | 1.5 GB | RTX 3060 12 GB |
| Qwen3-1.7B ↗ | 2.0B | BF16 | 4.1 GB | RTX 3060 12 GB |
| Qwen3-8B (above) ↗ | 8.2B | BF16 | 16.4 GB | 2× RTX 3060 12 GB |
| Qwen3-4B ↗ | 4.0B | BF16 | 8.0 GB | RTX 3060 12 GB |
From the model card
This is a clone of Qwen/Qwen3-0.6B with a multi-turn, tool-call compatible chat template.
Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.