Intel · Xe2 (Battlemage), 2x BMG-G21 (Xe2-HPG) · 48 GB VRAM tier
Best AI Models for Intel Arc Pro B60 Dual 48GB
Runs 14B and 35B MoE models well; 27–31B fits but is slow (13 tok/s).
48 GB is two 24 GB GPUs on one board, NOT a unified 48 GB pool — a single model must fit in 24 GB unless you use tensor/pipeline parallelism (vLLM, llama.cpp -ts).
| # | Model | Speed | llmrun Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.6 35B A3BAlibabaHow to run → Q4_K_M · 22.0 GB · 48.3 t/s | 48.3 tok/s | Q4_K_M · 22.0 GB | |
| 2 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 55.4 t/s | 55.4 tok/s | Q4_K_M · 16.5 GB | |
| 3 | Qwen3 30B A3B Thinking 2507Alibaba Q4_K_M · 19.4 GB · 53.7 t/s | 53.7 tok/s | Q4_K_M · 19.4 GB | |
| 4 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 61.9 t/s | 61.9 tok/s | Q4_K_M · 13.1 GB | |
| 5 | Q4_K_M · 19.3 GB · 59.7 t/s | 59.7 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Coding Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.6 35B A3BAlibabaHow to run → Q4_K_M · 22.0 GB · 48.3 t/s | 48.3 tok/s | Q4_K_M · 22.0 GB | |
| 2 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 55.4 t/s | 55.4 tok/s | Q4_K_M · 16.5 GB | |
| 3 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 61.9 t/s | 61.9 tok/s | Q4_K_M · 13.1 GB | |
| 4 | Q4_K_M · 7.9 GB · 43.0 t/s | 43.0 tok/s | Q4_K_M · 7.9 GB | |
| 5 | Q4_K_M · 19.3 GB · 59.7 t/s | 59.7 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Reasoning Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.6 35B A3BAlibabaHow to run → Q4_K_M · 22.0 GB · 48.3 t/s | 48.3 tok/s | Q4_K_M · 22.0 GB | |
| 2 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 55.4 t/s | 55.4 tok/s | Q4_K_M · 16.5 GB | |
| 3 | Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB · 35.8 t/s | 35.8 tok/s | Q4_K_M · 6.6 GB | |
| 4 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 61.9 t/s | 61.9 tok/s | Q4_K_M · 13.1 GB | |
| 5 | Q4_K_M · 19.3 GB · 59.7 t/s | 59.7 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Math Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.6 35B A3BAlibabaHow to run → Q4_K_M · 22.0 GB · 48.3 t/s | 48.3 tok/s | Q4_K_M · 22.0 GB | |
| 2 | Qwen3.8 27BAlibaba Q4_K_M · 17.9 GB · 13.1 t/s | 13.1 tok/s | Q4_K_M · 17.9 GB | |
| 3 | Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 13.1 t/s | 13.1 tok/s | Q4_K_M · 17.9 GB | |
| 4 | Q4_K_M · 19.3 GB · 59.7 t/s | 59.7 tok/s | Q4_K_M · 19.3 GB | |
| 5 | GLM 4.7 FlashZ.ai Q4_K_M · 19.9 GB · 45.0 t/s | 45.0 tok/s | Q4_K_M · 19.9 GB |
| # | Model | Speed ↓ | llmrun Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 69.9 t/s | 69.9 tok/s | Q4_K_M · 3.4 GB | |
| 2 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 61.9 t/s | 61.9 tok/s | Q4_K_M · 13.1 GB | |
| 3 | Q4_K_M · 19.3 GB · 59.7 t/s | 59.7 tok/s | Q4_K_M · 19.3 GB | |
| 4 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 55.4 t/s | 55.4 tok/s | Q4_K_M · 16.5 GB | |
| 5 | Qwen3 30B A3B Thinking 2507Alibaba Q4_K_M · 19.4 GB · 53.7 t/s | 53.7 tok/s | Q4_K_M · 19.4 GB |
| # | Model | Speed | Vision Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.6 35B A3BAlibabaHow to run → Q4_K_M · 22.0 GB · 48.3 t/s | 48.3 tok/s | Q4_K_M · 22.0 GB | |
| 2 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 55.4 t/s | 55.4 tok/s | Q4_K_M · 16.5 GB | |
| 3 | Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB · 35.8 t/s | 35.8 tok/s | Q4_K_M · 6.6 GB | |
| 4 | Gemma 3 12B ITGoogle Q4_K_M · 10.9 GB · 28.4 t/s | 28.4 tok/s | Q4_K_M · 10.9 GB | |
| 5 | MiniCPM V 4.7 35B A3BOpenBMB Q4_K_M · 22.1 GB · 53.4 t/s | 53.4 tok/s | Q4_K_M · 22.1 GB |
The Arc Pro B60 Dual 48GB runs 14B and 35B MoE models well. Xe2 (Battlemage), 2x BMG-G21 (Xe2-HPG) generation, released 2025.
Arc Pro B60 Dual 48GB strengths and weaknesses
Strengths
- Runs 35B MoE models: Qwen3.6 35B A3B at 48.3 tok/s in 22 GB
- Reads long prompts fast: A 4K-token prompt on Qwen3.6 35B A3B starts answering in 1.1 s
Weaknesses
- 120B MoE models don't fit: GPT OSS 120B needs 70.8 GB at Q4 vs 48 GB here
- Slow on large dense models: Llama 3.3 70B Instruct at 5.5 tok/s, Qwen3.8 27B at 13.1 tok/s
- oneAPI, not CUDA: No FP8 or NVFP4 weights and no vLLM AWQ or GPTQ; use GGUF builds
Output speed depends on memory bandwidth (456 GB/s). Grey rows need more than 48 GB.
- 9BQwen3.5 9B · Q4_K_M · 6.6 GB35.8tok/s
- 14BQwen3 14B · Q4_K_M · 11.9 GB24.0tok/s
- 27–31BQwen3.8 27B · Q4_K_M · 17.9 GB13.1tok/s
- 35B MoEQwen3.6 35B A3B · Q4_K_M · 22.0 GB48.3tok/s
- 5.5tok/s
- 120B MoEGPT OSS 120B · not at 4-bitneeds 70.8 GB
Estimated speeds: Q4_K_M, 4K context, one user. How we estimate
All 1216 models on the Arc Pro B60 Dual 48GB
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 |
|---|---|---|---|---|---|---|
Qwen3.5 4BVisionFunctions | Q4_K_M | 3.4 GB | 69.9 t/s | 262K | S86, Runs great | |
MiMo V2.6 Distill Qwen 9BVisionCodeFunctions | Q4_K_M | 6.5 GB | 36.7 t/s | 262K | A73, Runs well | |
Laguna XS 2.1ChatFunctions | Q4_K_M | 21.3 GB | 62.1 t/s | 262K | A84, Runs well | |
Hy Memory 30B A3BChatFunctions | Q4_K_M | 19.1 GB | 56.6 t/s | 262K | A82, Runs well | |
Laguna XS.2ChatFunctions | Q4_K_M | 21.3 GB | 62.1 t/s | 262K | A84, Runs well | |
Qwen3.5 9BVisionFunctions | Q4_K_M | 6.6 GB | 35.8 t/s | 262K | A73, Runs well | |
Spark X2.6 4BChatFunctions | Q4_K_M | 4.3 GB | 77.0 t/s | 1049K | S87, Runs great | |
Ornith 1.0 35BChatFunctions | Q4_K_M | 21.5 GB | 54.2 t/s | 262K | A81, Runs well | |
MiniCPM V 4.7 1BVisionFunctions | Q4_K_M | 1.6 GB | 178.1 t/s | 262K | S99, Runs great | |
LFM3 1.7BChatFunctions | Q4_K_M | 2.1 GB | 158.3 t/s | 262K | S97, Runs great | |
LFM3 Omni 2BVisionChatFunctions | Q4_K_M | 2.4 GB | 134.1 t/s | 128K | S95, Runs great | |
LFM3 VL 2BVisionChatFunctions | Q4_K_M | 2.3 GB | 147.1 t/s | 128K | S97, Runs great | |
ZAYA1 VL 8BVision | Q4_K_M | 7.5 GB | 100.1 t/s | 33K | S91, Runs great | |
Qwen3.6 35B A3B Abliterated V4ChatFunctions | Q4_K_M | 21.3 GB | 57.9 t/s | 262K | A83, Runs well | |
MiniCPM V 4.6VisionFunctions | Q4_K_M | 1.5 GB | 201.8 t/s | 262K | S100, Runs great |
Specs
- Architecture
- Xe2 (Battlemage), 2x BMG-G21 (Xe2-HPG)
- VRAM
- 48 GB GDDR6
- Memory bandwidth
- 456.0 GB/s
- FP16 (dense)
- 197.00 TFLOPS
- TDP
- 400 W
- Released
- 2025-08-01
- MSRP
- $1,200
Arc Pro B60 Dual 48GB FAQ
What is the largest model the Arc Pro B60 Dual 48GB can run?
Qwen2.5 72B Instruct (72.7B) is the largest at 4-bit or better: at Q4_K_S it uses 46.6 GB of 48 GB, and runs at about 5.4 tok/s.
Can the Arc Pro B60 Dual 48GB run Qwen3.6 35B A3B?
Yes. Qwen3.6 35B A3B fits at Q4_K_M, using 22.0 GB of 48 GB, and it is the best model that runs well on this card.
Can the Arc Pro B60 Dual 48GB run 120B MoE models?
No. GPT OSS 120B needs 70.8 GB at Q4, more than the 48 GB on this card.
How many tokens per second does the Arc Pro B60 Dual 48GB give?
About 48.3 tok/s on Qwen3.6 35B A3B at Q4_K_M, with the first token after 1.1 s on a 4K-token prompt.
What power supply does the Arc Pro B60 Dual 48GB need?
Plan a 800 W PSU (2× the 400 W TDP).
Is 48 GB enough for coding and agents?
Yes for coding: Qwen3.6 35B A3B, the best of the 4 coding models that fit, runs at about 48 tok/s. For agents, not yet: no agent-scored model fits in less than 96 GB.