ASUS’s October 31, 2026, Singapore listing puts the 64GB ProArt P16 H7607BA-SR409W at S$7,499, including GST, and marks it temporarily out of stock with back-order delivery by November 19. In the U.S., ASUS announced on October 16 that the ProArt P16 H7607BA was available online, with its listed starting MSRP at $2,799.99. Separately, DANAMIC reported short, AC-powered runs of three local AI workloads on a 64GB ProArt P16.
On this page
- U.S. pricing and the Singapore listing
- What DANAMIC ran locally
- Reported local AI results
- How far the results go
U.S. pricing and the Singapore listing
ASUS’s U.S. announcement lists a starting MSRP of $2,799.99 for the ProArt P16 H7607BA. The same announcement said online purchases were available through ASUS Store and Best Buy. That starting MSRP is for the P16 line, not a price for a particular 64GB configuration.
The Singapore listing applies to the exact SKU H7607BA-SR409W. On October 29, DANAMIC reported that ASUS had scheduled the Singapore sale for October 30. The later store snapshot, dated October 31, listed the model at S$7,499 including GST, temporarily out of stock, with back-order delivery by November 19. Those Singapore prices and order terms apply to that listing.
What DANAMIC ran locally
DANAMIC reported tests on an ASUS-loaned 64GB ProArt P16 running Windows 11 on Arm. The laptop was connected to AC power and set to Recommended performance mode in ProArt Creator Hub. For text generation, the review used Qwen2.5-7B Instruct in Q4_K_M format with a native Arm build of llama.cpp running on NVIDIA CUDA. Image and video generation used ComfyUI v0.42.1 with native Arm PyTorch and CUDA.
The results cover specific short runs, not a single shared performance score. The measured time ran from queuing a job to saving its output, so it includes some application overhead.
Reported local AI results
| Workload | Model and settings | Result reported by DANAMIC |
| Text generation | Qwen2.5-7B Instruct, Q4_K_M, llama.cpp on CUDA | About 54.6 tokens per second on GPU; about 51.3 tokens per second with 4,096 tokens of context; about 21.1 tokens per second on CPU alone. Prompt processing was about 3,000 tokens per second. |
| Image generation | Stable Diffusion XL (SDXL) base 1.0, 1,024 × 1,024 pixels, 20 steps | About 10.4 seconds for the first, cold image; an average of about 8.1 seconds across three warm images. |
| Video generation | Wan 2.1, 1.3-billion-parameter text-to-video model, 17 frames at 512 × 320 pixels, 12 steps | An average of about 7.5 seconds across two warm clips. |
The difference between the first cold SDXL image and the warm-image average belongs to that image test. The Qwen figures also distinguish GPU generation, generation with prior context, CPU-only generation and prompt processing; each measures a different part of the workload.
How far the results go
DANAMIC described the figures as a short baseline. Its review did not test sustained heavy workloads, AI processing on battery, NPU performance or output quality. The timings therefore describe the named runs under the reported AC-power conditions.
For Windows 11 on Arm, the review reports that some AI tools needed ARM64 builds, while standard model files did not require an Arm-specific download. That distinction matters when choosing software: check whether the application has an ARM64 build before setting up a workflow. NeoTeo previously covered RTX Spark tests across ProArt P16 workloads.
