MetaTrader 5 Build 5800 shipped in 2026 after the landmark Build 5572 that introduced native CUDA for ONNX. Here is what matters if you run ONNX models — and, just as important, what did not change so your existing EAs keep working.
TL;DR
Build 5800 expanded CUDA support for ONNX and fixed how the ONNX and OpenBLAS libraries are downloaded during updates. The OnnxCreate / OnnxRun API and the flags from 5572 are unchanged — no code changes required.
Expanded CUDA support for ONNX
Per the official Build 5800 release notes, MetaQuotes expanded CUDA support for ONNX models. In practice this is incremental over 5572: the flags and workflow you already set up (see what Build 5572 changed) still apply. After updating, re-confirm the GPU is actually engaged with these three checks.
The library-download fix (this one matters)
A frequent cause of ERROR 5800 was the ONNX runtime simply not being present. Build 5800 fixed updates for the ONNX and OpenBLAS libraries, with the required components now properly checked and downloaded during platform updates. If you ever hit "failed to load ONNX" because the runtime had not downloaded, updating to 5800 or later is part of the fix.
New OpenBLAS L1 trend-filtering
Build 5800 also added OpenBLAS methods for L1 trend filtering — fast, reproducible extraction of the trend component from a time series. That is a genuinely useful primitive for feature engineering: a clean trend feature computed natively in MQL5, without a round-trip to Python.
MetaEditor quality-of-life
- CSV files now open in a filterable, sortable table view — handy for inspecting exported features or trade logs.
- If your CPU supports AVX2, the editor auto-selects the matching compilation mode.
- Source files save as UTF-8 without BOM for MQL5 Algo Forge compatibility.
What did not change
The ONNX MQL5 API (OnnxCreate, OnnxRun, OnnxSetInputShape) and the 5572 flags (ONNX_USE_CPU_ONLY, ONNX_GPU_DEVICE_0..7, ONNX_ENABLE_PROFILING) are unchanged. Opset 17 remains the safe target for export. Existing EAs run without modification. Later builds (e.g. 5830) continued with general improvements.
Should you update?
Yes — especially if you ever saw the runtime fail to download. Update, then re-run your CUDA verification and a quick parity check to confirm nothing regressed.
Frequently asked questions
Does Build 5800 break my Build 5572 ONNX EA?
No. The ONNX MQL5 API and the 5572 flags are unchanged, so existing EAs run without modification. No code changes are required.
What is the ONNX change in Build 5800?
Expanded CUDA support for ONNX models plus a fix to how the ONNX and OpenBLAS libraries are checked and downloaded during platform updates, which resolves a common cause of ERROR 5800.
Running this live?
See which prop firms allow ONNX-driven EAs, or compare MT5 brokers for running an ML EA.