ONNX on MT5: Complete Guide to Running AI in Expert Advisors (Build 5572 + CUDA 2026)
Run ML models directly inside MT5 Expert Advisors without DLLs, sockets, or workarounds. This pillar covers everything: what ONNX is, why MQL5 supports it natively, the Python → ONNX → EA workflow, supported frameworks, honest limitations, and the Build 5572 updates: CUDA GPU support and new logging system.
🔥 Build 5572 Update (January 2026)
MT5 Build 5572 brought major changes to ONNX support: (1) Official CUDA GPU support — models run 5-15x faster on NVIDIA Turing+ GPUs. (2) Logging system reformulated — ONNX_DEBUG_LOGS is deprecated. Use: ONNX_LOGLEVEL_VERBOSE, ONNX_LOGLEVEL_INFO, ONNX_LOGLEVEL_WARNING, ONNX_LOGLEVEL_ERROR, ONNX_LOGLEVEL_FATAL. (3) On-demand loading — library loads only on first use. (4) CUDA settings in Tools → Options → Expert Advisors → Allow CUDA usage.
2026 pillar guide updated for MT5 Build 5572: native CUDA GPU support, new ONNX_LOGLEVEL_* logging system, Python -> ONNX -> EA workflow, supported frameworks and honest limitations.
🚀 To test ONNX EAs, free Deriv MT5 demo ($10,000 virtual):