Microsoft Executive Says RTX Spark Laptops Can Run 284B Models Locally

A Microsoft executive said RTX Spark laptops can run open-weight 284B AI models locally, including models that outperform OpenAI’s GPT-5 in coding and reasoning benchmarks.

The claim came during Microsoft’s Surface Laptop Ultra announcement. Pavan Davuluri, executive vice president of Windows + Devices, discussed the laptops’ ability to run DeepSeek-V4.1-Flash at 1.6-bit quantization entirely on-device, according to Wccftech.

The report describes the models as open-weight and says they can run locally on RTX Spark laptops. It attributes the performance comparison with GPT-5 to a Microsoft executive; it does not identify the benchmarks or provide results in the available account.

Separately, NVIDIA said during its RTX Spark presentation that laptops equipped with the SoC could run 120-billion-parameter models with a one-million Context Window. That figure was part of NVIDIA’s presentation and is distinct from the 284B model claim attributed to Davuluri.