A benchmark version number is specific to a test. Benchmark version numbers change rarely and only when absolutely necessary to accommodate changes in third-party applications or bug fixes.

Procyon AI Text Generation Benchmark for Windows v1.0.191.0

May 27, 2026

This release adds support for X2 Elite systems in the AI Text Generation benchmark. Following OpenVINO runtime updates in Procyon AI Benchmarks, performance characteristics may change across different intel hardware generations. We recommend benchmarking your devices again after an inference engine runtime update for up-to-date performance results.

New

  • Added support for Snapdragon X2 Elite systems 
  • Added NPU only (HTP) inference mode for Snapdragon X1 and X2 Elite systems while retaining the existing hybrid CNPU mode. The NPU mode is now the default in the GUI.  

Updated

  • Updated OpenVINO to 2026.1 
  • Updated QNN SDK to 2.43.1.260218. QNN models are generated with QNN SDK 2.40.280128
  • Updated the Genie runtime from using qai-appbuilder to C++ API

Fixed

  • Issues related to output token lengths for Genie and OV-NPU runtimes were fixed to ensure that the expected number of tokens are generated. 
  • Genie runtime for Snapdragons: The RAG chat templates used by Genie were slightly modified to generate sufficiently many tokens for the queries.
  • OpenVINO runtime for Intel NPUs: The compile configuration for RAG prompts now contains the following additional field: "MIN_RESPONSE_LEN": 512.  This change may cause cause scores to regress on certain platforms.  

Procyon AI Text Generation Benchmark for Windows v1.0.170.0

Mar 30, 2026


Fixed

  • Fixed an issue with offline benchmark runs where the benchmark attempted to download models even when they were already present on disk.

Procyon AI Text Generation Benchmark for Windows v1.0.152.0

Dec 19, 2025


New

  • Added support for Qualcomm Snapdragon X Elite devices using Genie. The first token runs on the NPU. The remaining inference happens on the CPU. 
  • Added support for Intel NPUS using OpenVINO. The entire workload runs on the NPU.
  • Added support for AMD XDNA 2 NPUs using ONNX Runtime combined with the VitisAI Execution Provider (VAIP). The benchmark offers both pure NPU mode and hybrid NPU + iGPU mode. In hybrid mode, the first token runs on the NPU while subsequent tokens are processed on the GPU. 

Updated

  • Updated system requirements to cover both GPUs and NPUs. 

Procyon AI Text Generation Benchmark for Windows v1.0.96.0

Aug 18, 2025

  • Updated OpenVINO to version 2025.2
    • This update brings significant performance improvements when run on Intel GPUs, with score improvements exceeding 50%.

Procyon AI Text Generation Benchmark for Windows v1.0.82.0

Mar 3, 2025

  • Updated OpenVINO to 2025.0. Please note that after this update, the first AI Text Generation Benchmark run using the OpenVINO runtime will take an extended amount of time, as the benchmark will delete AI models on disk and download models re-quantized with the latest runtime update tools.  
  • The ONNX Runtime inference engine has been updated to version 1.20.1 for the AI Text Generation Benchmark.
  • Fixed an issue where the Text Generation Benchmark would not run if the system was offline after models had been downloaded to the disk.  

Procyon AI Text Generation Benchmark for Windows v1.0.74.0

December 20, 2024

  • Fixed an issue with the AI Text Generation Benchmark failing when Procyon was installed on a drive other than “C:”  

Procyon AI Text Generation Benchmark for Windows v1.0.73.0

December 09, 2024

  • Launch version