Issues

• [CPU] ReduceSum loses int64 precision even when the result does not overflow

  microsoft/onnxruntime#33579 · Filter by author galleonli opened on Oct 25, 2026

• MatMulNBits benchmark generates IR12 models rejected by ORT 1.23.2

  model:transformer
  microsoft/onnxruntime#33576 · Filter by author arcusbuilds opened on Oct 24, 2026

• CoreML EP (ML Program): Conv without pads fails with "Required param 'pad' is missing"

  ep:CoreML
  microsoft/onnxruntime#33574 · Filter by author aktiwers opened on Oct 24, 2026

• [Web] Next release 1.26 is scheduled for May 2026 has passed

  platform:web
  microsoft/onnxruntime#33570 · Filter by author edumorlom opened on Oct 24, 2026

• [CoreML EP] Model cache keeps the ML Program package beside the compiled model, doubling its size

  ep:CoreML
  microsoft/onnxruntime#33566 · Filter by author reagle opened on Oct 23, 2026

• [CUDA EP] GroupQueryAttention with past/present bound to the same buffer gives nondeterministic, sometimes wrong results across runs (worse under CUDA graph capture)

  api:CSharpep:CUDA.NET
  microsoft/onnxruntime#33562 · Filter by author moshekar opened on Oct 23, 2026

• [WebNN] [WebGPU EP] [Windows ML] WebGPU EP via ExecutionProviderCatalog: wrong LibraryPath, and no way to require/upgrade to a minimum EP version

  ep:MIGraphXep:OpenVINOep:VitisAIep:WebGPUep:WebNNplatform:webplatform:windows
  microsoft/onnxruntime#33558 · Filter by author mingmingtasd opened on Oct 23, 2026
  1

• [WebGPU] Plugin EP terminates the host process when a Memcpy input fails to convert to a Tensor

  api:CSharpep:WebGPU.NETplatform:web
  microsoft/onnxruntime#33557 · Filter by author KakaruHayate opened on Oct 23, 2026
  3

• [Python] A requested execution provider that can't be created is silently dropped, e.g. a plugin MIGraphXExecutionProvider

  ep:MIGraphX
  microsoft/onnxruntime#33556 · Filter by author InfinitePortaldev opened on Oct 23, 2026

• [DirectML EP] MatMul with rank-5 inputs fails the whole session at initialization (80070057) instead of falling back

  ep:DMLep:WebNN.NET
  microsoft/onnxruntime#33549 · Filter by author mschneider85 opened on Oct 22, 2026

• CoreML EP (macOS arm64, 1.31.0): BNNS EXC_BAD_ACCESS in convolution_kernel for height-1 padded Conv inputs (VITS/piper graphs)

  ep:CoreML
  microsoft/onnxruntime#33544 · Filter by author willwade opened on Oct 22, 2026

• [Web] ort.min.js loads the JSEP wasm for executionProviders: ['wasm'], and WebKit then uses 5-7 GB (still in 1.30.0, follow-up to #26827)

  api:Javascript.NETplatform:mobileplatform:web
  microsoft/onnxruntime#33516 · Filter by author El1as42 opened on Oct 21, 2026
  5

• [Feature Request] Prebuilt onnxruntime-gpu TensorRT EP still requires TensorRT 10 libraries in TensorRT 11 environments

  ep:TensorRTfeature request
  microsoft/onnxruntime#33510 · Filter by author joe-nv opened on Oct 21, 2026
  1

• [Feature Request] Document that initializers added with data_is_external=true are read-only to ORT and EPs

  ep:OpenVINOep:WebGPUep:WebNNfeature requestmodel:transformerplatform:web
  microsoft/onnxruntime#33483 · Filter by author mingmingtasd opened on Oct 20, 2026
  2

• [Build] ORT 1.31.0: OpenVINO EP Python wheel fails to build (C2556 on TryGetProviderInfo_OpenVINO) — request 1.31.x patch release with #33327 backported

  buildep:OpenVINO
  microsoft/onnxruntime#33468 · Filter by author MayureshV1 opened on Oct 19, 2026

• [Feature Request] CUDA float16 support for GridSample

  ep:CUDA
  microsoft/onnxruntime#33455 · Filter by author mbenhamd opened on Oct 19, 2026

• [Bug][CPU] TransposeOptimizer inserts an unsupported INT32 QuantizeLinear after valid dequantization

  quantization
  microsoft/onnxruntime#33443 · Filter by author Yuhx141 opened on Oct 18, 2026

• [Bug][CPU] ReshapeFusion replaces a valid empty dynamic shape with an uninferable -1 dimension

  microsoft/onnxruntime#33442 · Filter by author Yuhx141 opened on Oct 18, 2026
  1

• [Bug][CPU] Dropping DequantizeLinear before ArgMax changes ordering for a finite negative scale

  quantization
  microsoft/onnxruntime#33440 · Filter by author Yuhx141 opened on Oct 18, 2026
  3

• [Bug][Training] TransposeReplacement creates an invalid Reshape for zero-sized dimensions

  microsoft/onnxruntime#33439 · Filter by author Yuhx141 opened on Oct 18, 2026

• [Bug][CPU] QLinearSoftmax lookup normalization assumes positive dequantization scale

  quantization
  microsoft/onnxruntime#33438 · Filter by author Yuhx141 opened on Oct 18, 2026
  1

• [Bug][CPU] QDQ MatMul fusion accepts INT32 activations unsupported by MatMulIntegerToFloat

  quantization
  microsoft/onnxruntime#33437 · Filter by author Yuhx141 opened on Oct 18, 2026
  1

• [Bug][CPU] QDQ Add fusion creates scalar-only QLinearAdd from legal rank-one quantization parameters

  quantization
  microsoft/onnxruntime#33436 · Filter by author Yuhx141 opened on Oct 18, 2026

• [Bug][CPU] NCHWc Add layout fallback creates an invalid reshape for a zero batch

  microsoft/onnxruntime#33435 · Filter by author Yuhx141 opened on Oct 18, 2026

• [CUDA EP] fp16 ReduceMean accumulates in fp16 and overflows to inf/NaN on large spatial reductions

  ep:CUDA
  microsoft/onnxruntime#33434 · Filter by author Iaotle opened on Oct 18, 2026