# LiteRT: what changed from 1 to 2

- Product: LiteRT (https://whatsnew.fyi/product/litert)
- Vendor: Google
- Range: changelog entries numbered after v1.4.1 up to and including v2.2.0, stable releases only
- Entries below: 9 releases (newest first)
- Resolved: 1 is v1.4.1 and 2 is v2.2.0, the newest stable release of each major we track
- Carrying security changes: 0 · CVEs mentioned: 0 · Mentioning breaking changes: 1 · Removing or deprecating something: 3
- Page: https://whatsnew.fyi/product/litert/compare/1...2

What's New is an index, not a publisher: every entry below links to the vendor's own release notes, which are the authoritative source. Entries are labelled where they are hand-curated sample data, pre-releases, or drawn from a secondary source such as a developer blog.

Reuse: the summaries, labels and curation here are © What's New. Quote freely with attribution and a link back; wholesale republication of the corpus is not permitted — terms: https://whatsnew.fyi/terms. The vendors' own release notes remain their publishers'.

## What changed (84 changes, grouped by kind)

### Added

#### v2.2.0 (2026-08-13)

- Published google-ai-edge-litert crate on crates.io enabling C++ SDK capabilities in Rust
- Added ABI version tracking to key modules (libLiteRT.so, GPU accelerator, Dispatch, Compiler) for smoother upgrades across 2.2.x binary versions
- Added LiteRT Tensor API support and features for recurrent KV cache swapping in LiteRT.js
- Added JIT/tracing support for custom tensor graphs in LiteRT.js
- Added native support for F8E4M3FN and F8E5M2 tensor element data types
- Added full JIT/AOT support for Intel NPU (LunarLake and PantherLake)
- Added new CPU accelerator YNNPACK that can be enabled through compilation and runtime flags
- Added ROPE composite op for GPU to improve LLM decode performance (~5% faster)
- Introduced new SetSelectedSignatures API
- Added RegisterCustomOpParser and MixinRegistrar to map custom flatbuffer operators to typed LiteRT operations
- Added FeedbackLoopConfig to automate recurrent KV cache swapping and zero-copy buffer bindings
- Added Tensor API python support with transformer operation surfaces, explicit Q/DQ export, and Safetensors loading
- Added JIT compilation and tracing support (compile() / WrittenModel) for custom tensor graphs in LiteRT.js
- Added integer overflow checks to convolution, reshape and pad-like CPU kernels

#### v2.1.6 (2026-07-02)

- Released ARMv7 prebuilts

#### v2.1.5 (2026-05-18)

- PyPI wheels for Python 3.14
- Support LiteRT environment and option APIs directly from Python
- Raspberry Pi 5 GPU acceleration support

#### v2.1.4 (2026-04-13)

- Introduced Environment to Python Compiled Model API to allow passing options

#### v2.1.3 (2026-03-17)

- Added experimental support for multi-threaded CompiledModel creation
- Added basic support for Broadcom VideoCore GPUs

#### v2.1.2 (2026-01-28)

- Desktop GPU support in the ai-edge-litert Python package for Linux through WebGPU and Mac through Metal
- Windows ai-edge-litert Python package with CPU inference support

#### v2.1.0 (2025-12-19)

- Custom op is supported through custom op dispatcher
- CMake Build is supported in addition to Bazel
- Released LiteRT C++ SDK using prebuilt libLiteRt.so file
- Added Profiler API in CompiledModel
- Added ErrorReporter API in CompiledModel
- Added ResizeInputTensor API in CompiledModel
- Introduced LiteRT Accelerator Test Suite for coverage and regression testing
- Introduced LiteRT graph transformation APIs for compiler plugins
- Added support for Qualcomm Snapdragon Gen5
- Added support for NPU JIT mode for Qualcomm and MediaTek
- Increased GPU coverage with WebGPU/Dawn and OpenCL including Android, Linux, MacOS, Windows, iOS, IoT devices
- Added asynchronous execution to Metal and WebGPU backends
- Added an option to control GPU inference priority
- Provided Desktop GPU backends prebuilt for Linux, MacOS, and Windows
- Added Interpreter API (CPU only) in Maven v2.1.0+ packages for Android

#### v2.0.3 (2025-11-13)

- Add Python backend for Google Tensor

### Changed

#### v2.2.0 (2026-08-13)

- Added new precision mode kLiteRtDelegatePrecisionFp16WithFp32Accum for GPU backend using FP16 storage and arithmetic with FP32 accumulation
- Changed batch size for GPU weight re-arrangement to scale with total model size for improved memory usage and initialization time
- Improved MTK NPU backend's channelwise quantization support for classical vision models
- Aligned Qualcomm QAIRT SoC coverage with SDK 2.47
- Added LiteRT.js API support for streamed weight loading Dynamic Range Quantized models to WebGPU

#### v2.1.6 (2026-07-02)

- Refactored CC APIs to be header only
- CC APIs can now be used without linking Abseil
- Expanded Accelerator Test Suite (ATS) single op coverage to 43 ops across f16 and f32 data types

#### v2.1.5 (2026-05-18)

- Make LiteRT C++ APIs header only
- Made LiteRT Options class as pure data only object without requiring a LiteRT C API call to create and update

#### v2.1.4 (2026-04-13)

- Improved c/cc API surfaces to avoid using deprecated methods
- Improved CMake build system's build stability

#### v2.1.3 (2026-03-17)

- Updated CMake build rules to support both CompiledModel and Interpreter APIs
- Expanded MediaTek NPU support to all applicable Android versions by supporting bundling MediaTek libraries in application binary
- Individual Options APIs (litert/cc/options) no longer use LiteRT C APIs (LiteRtXXX) and are updated to use string based serialization and use LrtXXX() functions which can be linked individually
- Move the experimental API GetProfiler() out of litert::CompiledModel
- Updated litert::Event::Type() to return C++ types instead of C types
- LiteRT headers no longer define OpenCL type names cl_mem and cl_event in the global namespace when OpenCL is not supported, replaced with type aliases LiteRtClMem and LiteRtClEvent in litert/c/litert_opencl_types.h
- LiteRT headers no longer define WebGPU type names struct WGPUBufferImpl and WGPUBuffer in the global namespace when WebGPU is supported, replaced with type alias LiteRtWGPUBuffer in litert/c/litert_webgpu_types.h

#### v2.1.0 (2025-12-19)

- Improved LiteRT Op coverage for Qualcomm and MediaTek
- Improved performance and memory footprint for GPU
- Improved memory utilization when executing on GPUs
- TensorBuffer::CreateManaged() requires Environment always
- All TensorBuffer creation requires Environment except HostMemory types
- LiteRT C++ constructors are hidden and all LiteRT C++ objects should be created by Create() methods
- Moved internal only C++ APIs to litert/cc/internal
- CompiledModel::Create() API no longer needs litert::Model and can be created from filename or model buffers directly

#### v2.0.3 (2025-11-13)

- Change manufacturer to Google and SoC models to include the Tensor_ prefix for Google Tensor
- Minor naming changes to some flags for the Google Tensor compiler plugin

### Fixed

#### v2.1.6 (2026-07-02)

- Improved built-in kernels to support high dimensional tensors and fixed GPU kernel issues like broadcast

#### v2.1.5 (2026-05-18)

- GPU prebuilts for iOS devices
- Bug on creating host memory TensorBuffer with alignment

#### v2.1.3 (2026-03-17)

- Fixed GPU support for Python on Windows
- Fixed a bug that given CPU Buffers are not always synced with GPU Accelerator from the second inference

#### v2.1.1 (2026-01-27)

- MacOS wheel build issue
- Qualcomm options passing issue in Kotlin API

#### v2.1.0 (2025-12-19)

- Fixes Android min SDK version to 23
- Fixes partition algorithm when the full model cannot be offloaded to NPU
- Better error handling without crashing on Delegation errors

### Removed

#### v2.1.5 (2026-05-18)

- libLiteRt.so dependency from GPU Accelerator and Dispatch API shared libraries

#### v2.1.3 (2026-03-17)

- Removed methods from litert::Event which uses C type LiteRtEnvironment

#### v2.1.0 (2025-12-19)

- Removed direct C headers usage
- Removed Tensor, Subgraph, Signature access from litert::Model
- Removed Annotation and Metrics APIs from CompiledModel
- Removed individual OpaqueOptions creation

## Release notes

### v2.2.0

- Date: 2026-08-13
- Version: v2.2.0
- Original notes: https://github.com/google-ai-edge/LiteRT/releases/tag/v2.2.0
- Permalink: https://whatsnew.fyi/product/litert/releases/v2.2.0

- **added** — Published google-ai-edge-litert crate on crates.io enabling C++ SDK capabilities in Rust
- **added** — Added ABI version tracking to key modules (libLiteRT.so, GPU accelerator, Dispatch, Compiler) for smoother upgrades across 2.2.x binary versions
- **added** — Added LiteRT Tensor API support and features for recurrent KV cache swapping in LiteRT.js
- **added** — Added JIT/tracing support for custom tensor graphs in LiteRT.js
- **added** — Added native support for F8E4M3FN and F8E5M2 tensor element data types
- **added** — Added full JIT/AOT support for Intel NPU (LunarLake and PantherLake)
- **added** — Added new CPU accelerator YNNPACK that can be enabled through compilation and runtime flags
- **added** — Added ROPE composite op for GPU to improve LLM decode performance (~5% faster)
- **added** — Introduced new SetSelectedSignatures API
- **added** — Added RegisterCustomOpParser and MixinRegistrar to map custom flatbuffer operators to typed LiteRT operations
- **added** — Added FeedbackLoopConfig to automate recurrent KV cache swapping and zero-copy buffer bindings
- **added** — Added Tensor API python support with transformer operation surfaces, explicit Q/DQ export, and Safetensors loading
- **added** — Added JIT compilation and tracing support (compile() / WrittenModel) for custom tensor graphs in LiteRT.js
- **added** — Added integer overflow checks to convolution, reshape and pad-like CPU kernels
- **changed** — Added new precision mode kLiteRtDelegatePrecisionFp16WithFp32Accum for GPU backend using FP16 storage and arithmetic with FP32 accumulation
- **changed** — Changed batch size for GPU weight re-arrangement to scale with total model size for improved memory usage and initialization time
- **changed** — Improved MTK NPU backend's channelwise quantization support for classical vision models
- **changed** — Aligned Qualcomm QAIRT SoC coverage with SDK 2.47
- **changed** — Added LiteRT.js API support for streamed weight loading Dynamic Range Quantized models to WebGPU

##### LiteRT Release Notes: Version 2.2.0
##### Executive Summary & Scope 

**API and Platform updates:**
* **Rust Integration** - Published google-ai-edge-litert crate on [crates.io](https://crates.io/crates/google-ai-edge-litert) enabling C++ SDK capabilities in Rust
* **ABI Compatibility** - Added ABI version tracking to key modules (libLiteRT.so, GPU accelerator, Dispatch, Compiler) facilitating smoother upgrades across 2.2.x binary versions.
**API Updates:** 
  * LiteRT.js - LiteRT Tensor API support and features for recurrent KV cache swapping; and JIT/tracing support for custom tensor graphs.
  * (check: do we call out SetSelectedSignatures?)
* **New Data Type** - Added native support for for F8E4M3FN and F8E5M2 tensor elements
**New HW support**
* Intel NPU - Added full JIT/AOT support for LunarLake and Panther Lake
* Qualcomm - QAIRT : Improved SoC coverage with SDK 2.47
* MediaTek  - Improvements to channelwise quantization support
* CPU (Experimental) : New YNNPack accelerator 
**Performance and Model optimizations**
* LLM throughput improvements - added ROPE composite Ops for GPU, resulting in ~5% faster decode performance
* Multiple optimizations for memory efficiency and GPU execution 
**Validation / Testing Infrastructure**
* Faster test registration 
* New multi-criteria validation framework (ConformanceSpec) for mixed-precision verification.


##### 🌟 Release 2.2.0 Summary
 


##### 1. LiteRT Runtime
Overview: Core C/C++ inference engine, delegate interfaces, platform APIs (Swift, Python, C), memory management, and hardware acceleration plugins.
* Old Revision: v2.1.6  
* New Revision: v2.2.0
* Repository: [google-ai-edge/LiteRT](https://github.com/google-ai-edge/LiteRT)
##### 🤖 Model/ Operator Coverage
XNNPACK supports for fp16 DEPTHWISE_CONV_2D, CONV_2D, FULLY_CONNECTED, and BATCH_MATMUL
⚡ Performance and Memory
* Added a new precision mode, kLiteRtDelegatePrecisionFp16WithFp32Accum.Currently it is GPU only. When enabled the GPU backend will use FP16 storage and arithmetic with FP32 accumulation for CONV_2D, DEPTHWISE_CONV_2D, FULLY_CONNECTED, TRANSPOSE_CONV and BATCH_MAT_MUL operators , which can provide a midground between pure 32-bit and pure 16-bit computation.
* Prepares input on GPU for the models which don’t run embedding lookup on CPU, including Gemma3 or before, and Open weight models like Qwen3
* Added ROPE composite op for GPU to improve the performance for LLMs.
The decode performance increased ~5%
* For best GPU performance, weights are re-arranged in batches on the GPU to be the optimal format for execution. In this release, the batch size was changed to scale with the total size of the model to reduce memory usage and improve initialization time. The impact will vary depending on model and device type but Gemma 4 E2B on a Linux NVIDIA machine saved ~200MB.
##### 🛠️ Hardware Acceleration
* [Experimental] New CPU accelerator YNNPACK can be enabled through the compilation flag: `--define litert_enable_ynnpack=true` and runtime flag: `--enable_ynnpack=true`. 
* Improved MTK NPU backend's channelwise quantization support for classical vision models, such as resnet18 and mobilenet.
* Added support for Intel NPU (LunarLake and PantherLake), enabling both JIT and AOT compilation. Verified on a variety of classic and selected GenAI models ([Gemma4 2B instructions](https://developers.google.com/edge/litert/next/litert_lm_npu#intel)) on Windows and Linux.
* LiteRT QAIRT accelerator
    * Major Features and Improvements
         - Supported specifying the Qualcomm SoC using either the SoC name or numeric SoC model.
         - Aligned SoC coverage with QAIRT SDK 2.47.
         - Refined OP validation and debugging logs.
         - Added documentation for QNN Native Path, QC options, and CMake toolchain builds.
         - Improved compiler plugin to avoid redundant QNN library reloads.
         - Enhanced OP coverage: A8W2 Fully Connected, BatchToSpaceNd

_[Truncated at 4000 characters — full notes: https://github.com/google-ai-edge/LiteRT/releases/tag/v2.2.0]_

### v2.1.6

- Date: 2026-07-02
- Version: v2.1.6
- Original notes: https://github.com/google-ai-edge/LiteRT/releases/tag/v2.1.6
- Permalink: https://whatsnew.fyi/product/litert/releases/v2.1.6

- **changed** — Refactored CC APIs to be header only
- **changed** — CC APIs can now be used without linking Abseil
- **added** — Released ARMv7 prebuilts
- **changed** — Expanded Accelerator Test Suite (ATS) single op coverage to 43 ops across f16 and f32 data types
- **fixed** — Improved built-in kernels to support high dimensional tensors and fixed GPU kernel issues like broadcast

#### Release 2.1.6


##### Major Features and Improvements

* Refactored CC APIs to be header only
* CC APIs now can be used without linking Abseil
* Released ARMv7 prebuilts
* Expanded Accelerator Test Suite (ATS) single op coverage to 43 ops across f16 and f32 data types.

##### Bug Fixes and Other Changes

* Improved built-in kernels to support high dimensional tensors, and fixed GPU kernel issues (like broadcast)

### v2.1.5

- Date: 2026-05-18
- Version: v2.1.5
- Original notes: https://github.com/google-ai-edge/LiteRT/releases/tag/v2.1.5
- Permalink: https://whatsnew.fyi/product/litert/releases/v2.1.5

- **added** — PyPI wheels for Python 3.14
- **added** — Support LiteRT environment and option APIs directly from Python
- **added** — Raspberry Pi 5 GPU acceleration support
- **changed** — Make LiteRT C++ APIs header only
- **changed** — Made LiteRT Options class as pure data only object without requiring a LiteRT C API call to create and update
- **removed** — libLiteRt.so dependency from GPU Accelerator and Dispatch API shared libraries
- **fixed** — GPU prebuilts for iOS devices
- **fixed** — Bug on creating host memory TensorBuffer with alignment

#### Release 2.1.5


##### Major Features and Improvements

* Python 3.14 Support: Added PyPI wheels for Python 3.14
* Make LiteRT C++ APIs header only
* Support LiteRT environment and option APIs directly from Python
* Removed libLiteRt.so dependency from GPU Accelerator and Dispatch API shared libraries. They no longer require libLiteRt.so to use.
   * Applied https://github.com/google-ai-edge/LiteRT/blob/main/g3doc/apis/runtime_context.md
* Made LiteRT Options class as pure data only object. It doesn’t require a LiteRT C API call to create & update the object.
* Added Raspberry Pi 5 GPU acceleration support

##### Bug Fixes and Other Changes

* Fixed GPU prebuilts for iOS devices
https://github.com/google-ai-edge/LiteRT/issues/6745 
* Fixed a bug on creating host memory TensorBuffer with alignment
https://github.com/google-ai-edge/LiteRT/issues/5373

### v2.1.4

- Date: 2026-04-13
- Version: v2.1.4
- Original notes: https://github.com/google-ai-edge/LiteRT/releases/tag/v2.1.4
- Permalink: https://whatsnew.fyi/product/litert/releases/v2.1.4

- **changed** — Improved c/cc API surfaces to avoid using deprecated methods
- **changed** — Improved CMake build system's build stability
- **added** — Introduced Environment to Python Compiled Model API to allow passing options

#### Release 2.1.4

##### Major Features and Improvements

* Improved [c/cc](http://c/cc) API surfaces to avoid using deprecated methods 
* Improved CMake build system’s build stability 
* Introduced Environment to Python Compiled Model API to allow passing options

### v2.1.3

- Date: 2026-03-17
- Version: v2.1.3
- Original notes: https://github.com/google-ai-edge/LiteRT/releases/tag/v2.1.3
- Permalink: https://whatsnew.fyi/product/litert/releases/v2.1.3

- **changed** — Updated CMake build rules to support both CompiledModel and Interpreter APIs
- **changed** — Expanded MediaTek NPU support to all applicable Android versions by supporting bundling MediaTek libraries in application binary
- **added** — Added experimental support for multi-threaded CompiledModel creation
- **fixed** — Fixed GPU support for Python on Windows
- **changed** — Individual Options APIs (litert/cc/options) no longer use LiteRT C APIs (LiteRtXXX) and are updated to use string based serialization and use LrtXXX() functions which can be linked individually
- **added** — Added basic support for Broadcom VideoCore GPUs
- **changed** — Move the experimental API GetProfiler() out of litert::CompiledModel
- **fixed** — Fixed a bug that given CPU Buffers are not always synced with GPU Accelerator from the second inference
- **removed** — Removed methods from litert::Event which uses C type LiteRtEnvironment
- **changed** — Updated litert::Event::Type() to return C++ types instead of C types
- **changed** — LiteRT headers no longer define OpenCL type names cl_mem and cl_event in the global namespace when OpenCL is not supported, replaced with type aliases LiteRtClMem and LiteRtClEvent in litert/c/litert_opencl_types.h
- **changed** — LiteRT headers no longer define WebGPU type names struct WGPUBufferImpl and WGPUBuffer in the global namespace when WebGPU is supported, replaced with type alias LiteRtWGPUBuffer in litert/c/litert_webgpu_types.h

#### Release 2.1.3


##### Major Features and Improvements

* Updated CMake build rules to support both CompiledModel and Interpreter APIs.
  cmake_example/CMakeLists.txt shows how you can use both libraries.

* Expanded MediaTek NPU support to all applicable Android versions
  (only Android 15 previously) by supporting bundling MediaTek libraries in
  application binary.

* Added experimental support for multi-threaded CompiledModel creation

* Fixed GPU support for Python on Windows

* Individual Options APIs (litert/cc/options) no longer use LiteRT C APIs (LiteRtXXX). They’re updated to use string based serialization and use LrtXXX() functions which can be linked individually.

* Added basic support for Broadcom VideoCore GPUs.


##### Bug Fixes and Other Changes

* Move the experimental API `GetProfiler()` out of `litert::CompiledModel`.

* Fixed a bug that given CPU Buffers are not always synced with GPU Accelerator
  from the second inference.

* Removed methods from `litert::Event` which uses C type `LiteRtEnvironment`.
  All C++ API should uses C++ `litert::Environment` instead.
  Also removed method `CreateFromSyncFenceFd()` that doesn't accept
  `litert::Environment`.

* Updated `litert::Event::Type()` to return C++ types instead of C types.

* The LiteRT headers no longer define the following OpenCL type names in the
  global namespace when OpenCL is _not_ supported: `cl_mem`, `cl_event`.
  These have been replaced with the type aliases `LiteRtClMem` and
  `LiteRtClEvent`, defined in a new header `litert/c/litert_opencl_types.h`.
  All of these symbols that didn't include the `LiteRt` prefix in their name
  were never intended to be part of the LiteRT API, and their presence
  in the global namespace risked conflicts with header files from other
  packages.

* Likewise, and for the same reason, the LiteRT headers no longer define the
  following WebGPU type names in the global namespace when WebGPU _is_
  supported: `struct WGPUBufferImpl`, `WGPUBuffer`. These have been
  replaced with the type alias `LiteRtWGPUBuffer` which is defined in a
  new header file `litert/c/litert_webgpu_types.h`. Alternatively, apps
  using these symbols can get them from WebGPU's `webgpu.h` header file.

* Added experimental support for multi-threaded CompiledModel creation

### v2.1.2

- Date: 2026-01-28
- Version: v2.1.2
- Original notes: https://github.com/google-ai-edge/LiteRT/releases/tag/v2.1.2
- Permalink: https://whatsnew.fyi/product/litert/releases/v2.1.2

- **added** — Desktop GPU support in the ai-edge-litert Python package for Linux through WebGPU and Mac through Metal
- **added** — Windows ai-edge-litert Python package with CPU inference support

#### Release 2.1.2

##### Major Features and Improvements

* Added Desktop GPU support in the ai-edge-litert Python package for Linux (through WebGPU) and Mac (through Metal).
* Released Windows ai-edge-litert Python package that supports CPU inference (WebGPU on Windows in Python is coming soon)

### v2.1.1

- Date: 2026-01-27
- Version: v2.1.1
- Original notes: https://github.com/google-ai-edge/LiteRT/releases/tag/v2.1.1
- Permalink: https://whatsnew.fyi/product/litert/releases/v2.1.1

- **fixed** — MacOS wheel build issue
- **fixed** — Qualcomm options passing issue in Kotlin API

#### Release 2.1.1

##### Bug Fixes and Other Changes

* Fixed the MacOS wheel build issue
* Fixed the Qualcomm options passing issue in Kotlin API

### v2.1.0

- Date: 2025-12-19
- Version: v2.1.0
- Original notes: https://github.com/google-ai-edge/LiteRT/releases/tag/v2.1.0
- Permalink: https://whatsnew.fyi/product/litert/releases/v2.1.0

- **added** — Custom op is supported through custom op dispatcher
- **added** — CMake Build is supported in addition to Bazel
- **added** — Released LiteRT C++ SDK using prebuilt libLiteRt.so file
- **added** — Added Profiler API in CompiledModel
- **added** — Added ErrorReporter API in CompiledModel
- **added** — Added ResizeInputTensor API in CompiledModel
- **added** — Introduced LiteRT Accelerator Test Suite for coverage and regression testing
- **added** — Introduced LiteRT graph transformation APIs for compiler plugins
- **added** — Added support for Qualcomm Snapdragon Gen5
- **added** — Added support for NPU JIT mode for Qualcomm and MediaTek
- **added** — Increased GPU coverage with WebGPU/Dawn and OpenCL including Android, Linux, MacOS, Windows, iOS, IoT devices
- **added** — Added asynchronous execution to Metal and WebGPU backends
- **added** — Added an option to control GPU inference priority
- **added** — Provided Desktop GPU backends prebuilt for Linux, MacOS, and Windows
- **added** — Added Interpreter API (CPU only) in Maven v2.1.0+ packages for Android
- **changed** — Improved LiteRT Op coverage for Qualcomm and MediaTek
- **changed** — Improved performance and memory footprint for GPU
- **changed** — Improved memory utilization when executing on GPUs
- **fixed** — Fixes Android min SDK version to 23
- **fixed** — Fixes partition algorithm when the full model cannot be offloaded to NPU
- **fixed** — Better error handling without crashing on Delegation errors
- **removed** — Removed direct C headers usage
- **removed** — Removed Tensor, Subgraph, Signature access from litert::Model
- **removed** — Removed Annotation and Metrics APIs from CompiledModel
- **removed** — Removed individual OpaqueOptions creation
- **changed** — TensorBuffer::CreateManaged() requires Environment always
- **changed** — All TensorBuffer creation requires Environment except HostMemory types
- **changed** — LiteRT C++ constructors are hidden and all LiteRT C++ objects should be created by Create() methods
- **changed** — Moved internal only C++ APIs to litert/cc/internal
- **changed** — CompiledModel::Create() API no longer needs litert::Model and can be created from filename or model buffers directly

#### Release 2.1.0

**Release 2.1.0 is the LiteRT** beta release.

LiteRT APIs are stable and have achieved feature parity. This milestone marks a significant step forward, introducing full feature parity with TensorFlow Lite, stable LiteRT APIs, and critical performance enhancements for GPU and NPU acceleration. With this release, we are officially recommending that developers begin their transition to LiteRT.

##### Major Features and Improvements

###### LiteRT Runtime

* Custom op is supported through [custom op dispatcher](https://github.com/google-ai-edge/LiteRT/blob/main/g3doc/apis/Custom_Op_Dispatcher.md).
* CMake Build is supported in addition to Bazel
* Released LiteRT C++ SDK using prebuilt libLiteRt.so file
* Added Profiler API in CompiledModel
* Added ErrorReporter API in CompiledModel
* Added ResizeInputTensor API in CompiledModel

###### LiteRT NPU

* Introduced LiteRT Accelerator Test Suite for coverage and regression testing
* Introduced LiteRT graph transformation APIs for compiler plugins
* Qualcomm
  * Added support for Qualcomm Snapdragon Gen5
  * Added support for NPU JIT mode
  * LiteRT Op coverage improvements
* MediaTek
  * Added support for NPU JIT mode
  * LiteRT Op coverage improvements

###### LiteRT GPU

* Increased GPU coverage with WebGPU/Dawn and OpenCL including Android, Linux, MacOS, Windows, iOS, IoT devices
* Added asynchronous execution to Metal, WebGPU backends
* Improved performance and memory footprint
* Added an option to control GPU inference priority
* Better error handling (without crashing) on Delegation errors

###### LLM Support

* Provided Desktop GPU backends prebuilt for Linux (x64, arm64), MacOS (arm64), Windows (x64)
* Improved memory utilization when executing on GPUs
* Published new LLMs on [https://huggingface.co/litert-community](https://huggingface.co/litert-community)
  * litert-community/FastVLM-0.5B
  * litert-community/Qwen3-0.6B
  * litert-community/embeddinggemma-300m with new NPU precompiled models
  * litert-community/gemma-3-270m-it with new NPU precompiled model
* Published Function Gemma on [https://huggingface.co/google](https://huggingface.co/google)
  * google/functiongemma-270m-it

###### LiteRT on Android

* Added Interpreter API (CPU only) in the Maven v2.1.0+ packages
* Added [Instruction](https://ai.google.dev/edge/litert/next/android_cpp_sdk) to use pre-built CompiledModell C++ API from the Maven package.

##### Bug Fixes and Other Changes

Fixes Android min SDK version and it’s 23 now.

LiteRT NPU: Fixes partition algorithm when the full model cannot be offloaded to NPU.

##### Breaking Changes

* Removed direct C headers usage. Users no longer need to include C headers.
* TensorBuffer::CreateManaged() requires Environment always.
* All TensorBuffer creation requires Environment except HostMemory types.
* LiteRT C++ constructors are hidden. All LiteRT C++ objects should be created by Create() methods.
* Moved internal only C++ APIs(such as litert\_logging.h) to litert/cc/internal
* Removed Tensor, Subgraph, Signature access from litert::Model. Instead users can access SimpleTensor, SimpleSignature.
* The CompiledModel::Create() API no longer needs litert::Model. They can be created from filename, model buffers directly.
* Users can access SimpleTensor and SimpleSignature from CompiledModel.
* Annotation, Metrics APIs are removed from CompiledModel.
* Removed individual OpaqueOptions creation. These OpaqueOptions objects are obtained by Options directly.
  *  Options::GetCpuOptions()
  *  Options::GetGpuOptions()
  *  Options::GetRuntimeOptions()
  * …

### v2.0.3

- Date: 2025-11-13
- Version: v2.0.3
- Original notes: https://github.com/google-ai-edge/LiteRT/releases/tag/v2.0.3
- Permalink: https://whatsnew.fyi/product/litert/releases/v2.0.3

- **added** — Add Python backend for Google Tensor
- **changed** — Change manufacturer to Google and SoC models to include the Tensor_ prefix for Google Tensor
- **changed** — Minor naming changes to some flags for the Google Tensor compiler plugin

#### Release 2.0.3

##### Major Features and Improvements

*  Add Python backend for Google Tensor. The backend doesn't yet register itself, so it's available by default.
* Change manufacturer to Google and SoC models to include the Tensor_ prefix for Google Tensor.
* Minor naming changes to some flags for the Google Tensor compiler plugin.

##### Bug Fixes and Other Changes
* N/A