Intel® oneAPI Toolkit Release Notes

Where to Find the Release

Intel® oneAPI Base Toolkit and Intel® oneAPI HPC Toolkit have now been unified into Intel® oneAPI Toolkit with the 2026.0 Release. You can get the latest version of Intel® oneAPI Toolkit from here.

If you have Priority Support for Intel® oneAPI Toolkit or any other oneAPI toolkit (i.e. an unexpired free license or a paid license with active support at the time of the build date for this product release), you can download the installer by logging in to Intel® Registration Center and selecting the appropriate product. You may need to create an account and/or register your product. For additional information, please see: Product Registration and Sign-Up FAQ

Customers of Intel® oneAPI Base Toolkit and Intel® oneAPI HPC Toolkit will be automatically transitioned and upgraded to Intel® oneAPI Toolkit in their Intel® Registration Center accounts and can download older versions of the Base kit and HPC kit from their accounts as well.

Intel® oneAPI Toolkit Component Versions

Please visit Intel® oneAPI Toolkit and Component Versioning Schema for semantic versioning schema detail. 

The following table contains a mapping of component versions in each Intel® oneAPI Toolkit release

Note: Intel® Advisor has been removed from the Intel® oneAPI toolkit and is now available as a stand-alone download here. This release includes security and functional bug fixes. Starting with the 2026.0 release, Advisor was decoupled from oneAPI and delivered as a stand-alone product, aligned with oneAPI release schedules.

Intel® oneAPI Data Analytics Library (oneDAL) and Intel® oneAPI Deep Neural Network Library (oneDNN) are not included in the Intel® oneAPI Toolkit installer package but the compatible versions are available as stand-alone release for separate download 

New in this Release

2026.2

Highlights

• Get more performance from the latest Intel® CPUs and GPUs through optimizations built into the latest Intel® oneAPI Toolkit 2026.2 release. Enhancements across compilers, libraries, and performance tools take advantage of capabilities such as Intel® AVX10, Intel® AMX, and next-generation Intel graphics architectures, reducing the amount of manual platform-specific tuning required for your applications.
• Run larger AI, analytics, and HPC workloads more efficiently through enhancements in key performance libraries including Intel® oneMKL, Intel® oneDAL, Intel® oneDNN, Intel® MPI Library, and Intel® oneCCL. Improvements in numerical kernels, distributed analytics, communication runtimes, and collective operations help accelerate common workloads while improving scalability across CPUs, GPUs, and clusters.
• Reduce overhead in CPU and GPU applications through improvements in SYCL runtimes, communication libraries, and memory management capabilities. New capabilities help move data more efficiently, improve graph execution, and reduce the performance costs often associated with heterogeneous programming.
• Adopt the latest C++, Fortran, SYCL, and OpenMP capabilities more easily through expanded standards support, faster SYCL build workflows, and new language features that help modernize and optimize existing applications without large-scale rewrites.
• Spend less time finding and fixing performance bottlenecks with enhanced profiling, debugging, and analysis capabilities in Intel® VTune™ Profiler and Intel® Distribution for GDB*. New AI-assisted workflows, expanded GPU visibility, and support for the latest Intel platforms help developers identify optimization opportunities more quickly.
• Scale applications more effectively from workstations to distributed clusters through enhancements in Intel® MPI Library, Intel® oneCCL, Intel® oneDAL, and Intel® oneTBB that improve workload distribution, communication efficiency, and resource utilization.

Intel® oneAPI DPC++/C++ Compiler

• Accelerate SYCL build times by pre-compiling large C++ and SYCL headers once for both host and device targets. This unified precompiled header support significantly speeds up iterative edit-build-debug cycles, boosting developer productivity on complex codebases.
• Enable zero-copy access and accelerate data movement between the CPU and GPU without rewriting legacy memory allocators. By registering standard host memory directly with the SYCL runtime via sycl_ext_oneapi_register_host_memory, developers can achieve high-performance data access on supported Intel® GPUs using the Level Zero backend, minimizing application refactoring.
• Build highly dynamic and extensible GPU-accelerated applications by linking pre-compiled and runtime-generated kernels. The compiler seamlessly fuses ahead-of-time compiled SYCLBIN objects with JIT-compiled source bundles, empowering developers to create sophisticated plugin architectures, runtime code generators, and dynamic AI workloads.
• Leverage cutting-edge OpenMP 5 and 6 features to extract maximum performance from modern multicore and heterogeneous systems. This release enhances support for advanced tasking with transparent dependencies, maximizes core utilization with threadset scheduling, and provides robust adjust_args modifiers for safer, more flexible GPU offloading.

Intel® Fortran Compiler

• Scale complex parallel simulations across all CPU cores and prevent GPU offload crashes using key OpenMP 6.0 and 6.1 features. The transparent and threadset clauses enable dynamic, cross-generational task scheduling to keep all cores saturated, while new adjust_args fallback modifiers ensure unmapped host pointers are handled gracefully in GPU variants without requiring developers to maintain duplicate code paths.
• Write modern, reusable scientific libraries without the maintenance overhead of duplicating code for different precisions. New Fortran 2023 TYPEOF and CLASSOF statements allow your generic routines, interfaces, and preprocessor macros to automatically inherit the type and precision from calling variables, effectively eliminating the need to copy-paste and manage separate REAL4, REAL8, and COMPLEX versions of the same algorithm.
• Protect your simulations from subtle data corruption bugs caused by string handling without adding manual buffer management code. The compiler's support for Fortran 2023 automatic reallocation of deferred-length character variables ensures that strings passed to intrinsic procedures are precisely resized to fit return values, preventing silent data truncation and eliminating a common source of corrupted file paths and output logs.

Intel® oneAPI DPC++ Library

• Developers can now take advantage of 14 newly added range algorithms in the DPC++ Library, including capabilities for sorting, partitioning, heap validation, rotation, and data movement operations. This addition completes support for all range algorithms defined in the oneDPL v1.5 specification, giving developers a richer, standards-aligned toolkit to build modern C++ applications with greater productivity and portability.
• Reduced data transfer overhead for sycl::buffer based algorithms by eliminating unnecessary host-to-device copies, improving efficiency for heterogeneous applications.
• Faster execution of scan-based algorithms on CPUs through enhancements via the oneTBB backend, accelerating common operations such as copy, remove, and unique.

Intel® VTune™ Profiler

• Support for the latest Intel® Core™ Ultra processors across CPU, GPU, and XPU analysis — tune for the next generation of hardware as soon as you get the parts, using the same workflows you already know.
• Drive VTune Profiler with the new AI Assistant. A built-in AI agent and MCP server let you discover your VTune installation, pick the right analysis, run collections and reports, and get bottleneck insights with suggested code changes — conversationally, with your own LLM, from your own environment.
• New GPU Hotspots & XPU Offload features: 
  • See the real cost of your SYCL graphs. Support for the zeCommandListImmediateAppendCommandListsExp Level Zero API means SYCL graph workloads (native graphs excluded) are now traced correctly in GPU Hotspots and XPU Offload — no more missing or misattributed GPU tasks in graph-based AI and HPC pipelines.
  • Diagnose GPU occupancy limits at a glance. GPU Hotspots and XPU Offload now collect and display General Register File (GRF) size for OpenCL™ and Level Zero kernels in a dedicated column.

Intel® oneAPI Math Kernel Library

• Faster Random Number Generators: Run larger Monte Carlo studies in the same simulation window on your newest Intel® CPUs. Random Number Generators are now tuned for Intel® AVX10 instruction set support, delivering higher sample generation throughput on capable Intel® CPUs — directly shortening the runtime of the variate-hungry workloads at the core of risk analytics, option pricing, particle transport, and stochastic engineering models, with no changes to the generator APIs or the statistical properties of the streams.
• Added fp16 support for Sparse BLAS SYCL APIs: Cut memory traffic in sparse GPU workloads by moving them to half precision where the accuracy budget allows. Selected Sparse BLAS SYCL routines now accept half (fp16) data, so sparse matrix operations can hold twice as much of the matrix in the same bandwidth and cache footprint — running in the same SYCL code path as your existing single- and double-precision calls, on the routines where fp16 support has landed.
• Library quality improvements: Floating point exceptions have been significantly reduced when building with strict floating point exceptions option.

Intel® oneAPI Threading Building Blocks

• Introduced production support for the task arena leave_policy and a new global_control default parameter. For developers building multi-tenant or hybrid-runtime applications, this update eliminates idle CPU spinning by letting worker threads exit an arena immediately once work is depleted. It allows you to dynamically manage thread lifetimes across your entire application, preventing thread oversubscription and reclaiming critical CPU cycles when transitioning between competing threading runtimes.
• Production support for the parallel_phase API, which marks a region where worker threads are retained in a task arena longer, so the arena stays reactive to new parallel work instead of re-recruiting threads for each burst. A phase can specify an optional leave_policy to apply when the region ends.
• A new starvation avoidance mechanism in the resource_limited_node preview feature, improving fairness among nodes that compete for the same limited resource and reducing average message latency in those flow graphs.

Intel® oneAPI Collective Communications Library

• Smarter, topology-aware communication for both scale-up and scale-out with automatic PCIe switch discovery and topology-aware sub-communicators (node/NUMA/switch/none). 
• Major performance improvements across key collectives, including one-shot, OFI-direct, and proxy-thread-based optimizations for reduce-scatter, allgather, alltoall, and allreduce in both scale-up and scale-out deployments. Additional enhancements include optimized broadcast, new alltoallv and scatter collectives, optimized point-to-point communication, low-latency SYCL reduce, AVG reduction, and custom pre-op support. Introducing support for native graph recording further boosts scale-up performance.
• oneCCL now defaults to OFI/libfabric instead of MPI, adds memory window registration API (LSA) for direct load/store access, and strengthens multi-threading support by introducing a new communicator split API and various enhancements.

Intel® oneAPI Data Analytics Library

• Added the option to opt into AMX BF16 dispatch for Euclidean distance in KNN and DBSCAN workloads to unlock significant performance gains on modern Intel hardware.
• Expanded clustering capabilities with oneDAL's new HDBSCAN algorithm and distributed Ridge Regression. Tackle complex, variable-density datasets with HDBSCAN, while Ridge Regression now supports distributed SPMD/MPI CPU training, enabling faster, scalable model building across large datasets and multi-node environments.
• Preprocess your data faster with the new hardware-accelerated MaxAbsScaler in sklearnex. This drop-in estimator helps you reduce preprocessing bottlenecks and move from raw data to trained models more efficiently.

Intel® Distribution for GDB*

• Added debugging support for Intel® Core™ Ultra processors.
• Added support for Microsoft Visual Studio* 2026.
• Bug Fixes & Improvements: Includes critical bug fixes and functional updates to improve stability and reliability.

Intel® oneAPI Deep Neural Network Library

• Next-gen CPU performance with Intel® Core™ Ultra processors: Customers targeting future Intel® Core™ Ultra processors gain optimized performance through Intel® AVX10.2 instruction set support, delivering faster AI and compute workloads out-of-the-box, minimizing optimization effort and accelerating time-to-production with no additional code changes required.

• Next-gen integrated GPU readiness with future Intel® Graphics: Customers developing GPU-accelerated workloads gain early optimized support for future Intel® integrated GPUs, ensuring applications are performance-ready at hardware launch.

• Improved support for SYCL Graph: Customers can now leverage SYCL Graph native recording mode with oneDNN on Intel® GPUs, enabling automatic capture and replay of GPU command sequences and reducing CPU overhead and synchronization costs, resulting in more efficient, high-throughput GPU execution pipelines for demanding AI and compute-intensive applications.

Intel® Integrated Performance Primitives Library

• Process longer signal streams without splitting them into chunks to dodge 32-bit index limits. Intel IPP continues to extend its Signal Processing domain with 64-bit size support, adding 64-bit variants to a growing set of functions so they can address arrays beyond the previous 32-bit element bounds — letting audio, radar, and sensor pipelines drop the buffer-tiling and offset-bookkeeping workarounds wherever a 64-bit variant is available.
• Get more throughput from the AVX-512 servers you already own, with no code changes at all. This release delivers broad performance improvements across IPP image processing, computer vision, and data compression functions on AVX-512 hardware, so existing applications speed up simply by relinking — accelerating image pipelines, vision inference preprocessing, and compression-heavy storage and transfer workloads without touching a line of source.
• Push high-volume image preprocessing through in a single call instead of paying per-image setup costs thousands of times over. New batch processing APIs for generic image resize and warp transformations let applications submit many images — or many tiles of one large image — as one operation, amortizing initialization and dispatch overhead across the whole set.

Intel® Cryptography Primitives Library

• Optimized ML-KEM key encapsulation and decapsulation specifically targeting Intel® AVX-512 instruction sets on Intel® Xeon® CPUs. For infrastructure and security engineers tasked with future-proofing data-in-transit, this optimization eliminates the processing bottlenecks of post-quantum handshakes. It allows you to run quantum-resistant TLS and VPN services at silicon speed, keeping your latency SLAs intact without driving up your hardware or cloud hosting costs.
• Introduced a new external Mu verification API and built-in FIPS self-tests for ML-DSA operations, covering key generation, signing, and verification. For embedded and compliance engineers tasked with securing resource-constrained or regulated devices, this update enables streaming verification of multi-gigabyte firmware images without exhausting system memory. It allows you to offload heavy hashing computations to separate hardware modules and instantly validates your cryptographic pipeline with out-of-the-box self-tests, keeping your hardware efficiency high while accelerating your FIPS compliance timeline.

Intel® MPI Library

• Libfabric goes standalone in oneAPI: Developers can now install and update libfabric independently of the full oneAPI toolkit, enabling faster access to fabric fixes and features on their own timeline and removing legacy PSM2/PSM3 providers to fully adopt modern rxm/ verbs, and OPX providers
• Added native XPMEM (Cross Partition Memory) support, closing the performance gap with OpenMPI on XPMEM-enabled clusters. New XPMEM support delivers faster intra-node, shared-memory communication on XPMEM-enabled systems for latency-sensitive HPC and AI workloads.
• Implemented targeted performance tuning for both single-node and multi-node all-to-all collective communication algorithms to boost performance and scalability for communication-heavy workloads like MoE models and distributed FFTs.
• Broader OS support: Intel MPI now supports Ubuntu 24.04 for both single-node and multi-node deployments, giving customers more flexibility in choosing their Linux distribution.
• Introduced automatic platform recognition and high-speed shared-memory (shm) transport support specifically for the upcoming Intel platforms.
• Thread-Split Support for UCX Provider: Implemented thread-split support specifically for the UCX (Unified Communication X) network provider, which allows multiple threads to concurrently utilize UCX communication resources without lock contention.

2026.1

Highlights

• Intel® oneAPI Toolkit 2026.1 improves performance and scaling of AI and HPC workloads through targeted optimizations across libraries and runtimes, including faster math routines and more efficient communication, allowing applications to run faster on Intel CPUs and GPUs without code changes.

• Developers can more effectively target next-generation Intel GPUs with expanded platform support, including initial enablement for Intel Xe3-based GPUs and deeper hardware awareness that helps map workloads more efficiently to GPU architecture.

• AI and data analytics workflows integrate more easily with improved ecosystem support, including compatibility with newer sklearn versions and more consistent APIs across algorithms, reducing friction when adopting Intel® oneAPI in Python environments.

• Applications scale more efficiently across nodes and devices with enhancements such as GPU RDMA support and improved collective communication behavior, helping distributed and multi-GPU workloads achieve better performance and reliability.

• High-performance math and data processing are further optimized with Intel® oneAPI Math Kernel Library improvements, such as faster LAPACK routines on GPUs, enabling applications to achieve better performance without code modifications.

• Developers can write and optimize parallel code more productively with more parallel range algorithms, improved NUMA-aware memory control, and emerging C++20 module support that can reduce build times and improve code organization.

• Performance tuning is more effective with deeper insight into accelerator behavior, including visibility into power and execution characteristics of emerging compute engines, enabling better optimization of heterogeneous workloads.

Intel® oneAPI DPC++/C++ Compiler

• The Intel® oneAPI DPC++/C++ Compiler reduces runtime overhead and improves scaling by capturing mixed SYCL and Level Zero workloads into a single execution graph, which is critical for AI frameworks and complex HPC applications.

• Large applications start faster and run more efficiently due to improved kernel query performance, reducing overhead in codes with many kernels.

• Developers can better tune performance on Intel Xe GPUs by using detailed hardware topology insights such as tiles, cores, and execution engines to more effectively map workloads.

• The compiler enables tighter integration with AI frameworks and custom runtimes by allowing host callbacks to be inserted directly into execution streams without breaking SYCL execution.

• Improved handling of SYCL kernels and lambdas reduces build friction and makes development more predictable across environments.

Intel® Fortran Compiler

• The Intel® Fortran Compiler makes GPU offload easier to adopt by allowing developers to enable unified shared memory across an application with a compiler option, so existing OpenMP and DO CONCURRENT code can run on GPUs without code changes.

• The Fortran compiler support of Unified Shared Memory reduces the need for manual data management, making it significantly easier to move Fortran applications between CPUs and GPUs and accelerate modernization of existing HPC code.

• Developers can better tune performance of parallel workloads by controlling how OpenMP and DO CONCURRENT loops execute across heterogeneous systems.

• Improvements in OpenMP offload efficiency help existing Fortran applications make better use of GPU resources with minimal effort.

• Applications behave more predictably at scale as improvements in OpenMP and loop execution increase reliability for complex, long-running HPC workloads.

Intel® oneAPI DPC++ Library

• Comprehensive Algorithm Expansion: Added a robust suite of parallel range algorithms featuring copy operations (replace_copy, remove_copy), advanced search capabilities (find_last variants), and powerful comparison functions (starts_with, ends_with, contains, lexicographical_compare).

Intel® VTune™ Profiler

• XPU Offload Analysis: When NPU profiling is enabled, data collection now includes NPU power trace data, giving developers deeper visibility into performance and power behavior for AI workloads.

• Bug Fixes & Security: Includes critical bug fixes and security updates to improve stability and reliability.

Intel® oneAPI Math Kernel Library

• Extended platform support: Introduced initial support for Intel® GPUs that are based on Intel® Xe3P microarchitecture.

• More optimization: Improved the performance of lapack::orgqr for all architectures of Intel® GPUs.

Intel® oneAPI Threading Building Blocks

• NUMA Optimization: Use new interleaved memory allocation APIs to explicitly control memory placement across NUMA nodes. The allocate_numa_interleaved and deallocate_numa_interleaved functions enable allocating memory across multiple NUMA nodes, helping increase bandwidth and avoid bottlenecks from single-node allocation.

• C++20 Modules: This release adds experimental support for C++20 modules with the introduction of include/oneapi/tbb.cppm. Developers can use this instead of traditional headers to reduce compilation times and improve encapsulation.

Intel® oneAPI Collective Communications Library

• Performance Improvements: Multi-segment memory support enables more efficient IPC exchanges for large-scale applications. Enhanced alltoall operations with pre-copy size validation prevent errors and improve reliability. Custom reductions optimized specifically for Battlemage architecture deliver better performance

• Reliability Enhancements: Improved concurrent insertion handling for shared resources reduces conflicts in multi-threaded environments. Fixed event handling ensures accurate runtime state tracking. Enhanced allreduce and reduce-scatter operations with proper custom pre-operation enforcement

• GPU & Communication Optimizations: GPU_RDMA integration for point-to-point operations when hardware supports it. Deterministic scaleout host buffer allocation improves consistency. Simple protocol now properly handles unaligned data counts for broader compatibility

Intel® oneAPI Data Analytics Library

• Intel® Extension for Scikit-learn* introduces support for Sklearn 1.9 and implemented the "everything follows X" design pattern across linear models, KNN, decision forests, and basic statistics, providing users with more consistent APIs and broader framework compatibility for seamless integration.

• Added SPMD (Single Program, Multiple Data) API support for PCA and Covariance computations on CPU, enabling faster processing of large datasets through improved parallel execution and better resource utilization.

• Enhanced SVM models with iteration tracking and added comprehensive namespace/device checks for logistic regression, giving data scientists better insights into training progress and reliable model deployment across different computing environments.

Intel® Distribution for GDB*

• Bug Fixes and Security: Includes updates for critical functional bug fixes and security issues.

2026.0

Highlights

• With the 2026.0 release, Intel® oneAPI Toolkit brings together the former Intel® oneAPI Base Toolkit and Intel® oneAPI HPC Toolkit, providing the same trusted compilers, libraries, and analysis tools developers rely on today, now delivered as a single toolkit for building performance‑tuned AI, client, edge, and HPC applications.
• Intel® oneAPI Toolkit continues to enable cross‑architecture CPU, GPU, and accelerator development, and in 2026.0 adds support for Clearwater Forest and Intel® Core™ Series 3 processor while building on optimizations for Arrow Lake and Intel® Core™ Ultra systems and strengthening Intel® Arc™ B‑series GPU development (including Intel® Arc™ Pro B70 Graphics) through improved code generation, runtime behavior, and profiling.

• The Intel® oneAPI DPC++/C++ Compiler 2026.0 helps developers deliver higher‑performance heterogeneous C++/SYCL* applications faster by boosting execution efficiency through improved code generation (including better branch prediction) on Intel® Core™ Ultra and Arrow Lake platforms, providing strong readiness for Clearwater Forest and Intel® Core™ Series 3 processor, cutting iteration time with quicker AOT linking, making SYCL code more expressive and portable with enhanced free‑function kernels, device‑type queries, and expanded SYCL Graph execution using work‑group scratch memory, and making tuning and debugging more targeted with function‑scoped optimization and offload reporting, loop‑optimization controls, and more robust host‑and‑offload execution behavior.

• The Intel® Fortran Compiler 2026.0 is ready to boost performance on Clearwater Forest processors and modernizes Fortran development by adopting the standard dispatch construct for cleaner, more portable offload code, improving numerical output predictability with leading‑zero editing, enabling more scalable parallel coordination with PUT with NOTIFY for coarrays, making code clearer with conditional expressions, producing higher‑quality generated code via extended prefer_type IR support, and reducing manual data‑mapping work for heterogeneous applications with automap support for allocatables on declare target enter.

• Intel® oneAPI Math Kernel Library and Intel® Integrated Performance Primitives expand optimized coverage and scalability for core math, signal‑processing, and data‑processing kernels in 2026.0, improving out‑of‑the‑box performance for simulation, rendering, media, and analytics workloads as CPU core counts and memory hierarchies continue to grow.

• Intel® Integrated Performance Primitives in 2026.0 further enhance performance for signal processing, media, and data‑processing workloads, including enhanced Intel® Cryptography Primitives Library which accelerates security‑sensitive and data‑intensive applications.

• Intel® oneAPI Deep Neural Network Library and Intel® oneAPI Data Analytics Library strengthen support for machine learning, deep learning, and data‑centric workloads, enabling more efficient training, inference, and preprocessing pipelines that integrate cleanly with modern AI and analytics frameworks on CPUs and GPUs.
• Intel® oneAPI Threading Building Blocks and Intel® oneAPI DPC++ Library enhance task‑based parallelism and standard‑based parallel algorithms, helping client, edge, and workstation applications scale efficiently across cores without manual thread management.
• As applications extend beyond a single system, the Intel® MPI Library and Intel® oneAPI Collective Communications Library provide scalable communication and coordination for multi‑node HPC simulations, distributed AI training, and large‑scale analytics workloads.
• Intel® VTune™ Profiler in 2026.0 delivers deeper, more actionable insight into compute performance, memory behavior, vectorization, threading, GPU execution, and communication, enabling developers to identify and tune performance issues across client, edge, and high‑performance environments using a unified analysis approach.

Intel® oneAPI DPC++/C++ Compiler

• Faster build and iteration cycles through runtime performance improvements, including quicker AOT linking, helping developers scale large applications with less compile and link overhead.

• More performance on next‑generation Intel® platforms with improved branch prediction code generation for Intel® Core™ Ultra Series 3 processor and Arrow Lake CPUs, delivering higher execution efficiency without requiring code changes.

• More expressive and portable SYCL development with enhanced free function kernel support, new device‑type query extensions, and expanded SYCL Graph capabilities using work‑group scratch memory—enabling cleaner code and more efficient execution across devices.

• Clearer insight into compiler optimizations and offload behavior with function‑scoped optimization reporting, plus new reports for reduction optimizations and ND‑range partitioning, making performance tuning more targeted and actionable.

• Greater control during performance tuning and debugging through new compiler options that allow developers to explicitly enable or disable loop optimizations when isolating issues or experimenting with transformations.

• More predictable and robust host and offload execution with added fallback controls for need_device_ptr handling and new scan support for scalars in non‑SIMD host parallel loops, improving correctness across heterogeneous workloads.

• A cleaner, more future‑ready programming model with removal of deprecated target variant dispatch and subdevice support, simplifying development and aligning with modern SYCL and offload best practices.

Intel® Fortran Compiler

• A cleaner and more future‑proof parallel programming model by removing the legacy target variant dispatch construct in favor of the standard dispatch construct, simplifying offload code and improving long‑term portability and maintainability.

• More precise and standards‑compliant formatted output with support for leading‑zero editing, helping developers produce clearer, more predictable numeric output for scientific, engineering, and reporting workloads.

• Improved synchronization and coordination in parallel programs with support for PUT with NOTIFY, enabling more efficient and expressive communication patterns in distributed and coarray‑based applications.

• More expressive and readable Fortran code through support for conditional expressions, allowing developers to write clearer logic with fewer branches while maintaining performance.

• Stronger optimization and transformation opportunities inside the compiler with extended prefer_type IR support for implementation attributes, enabling better internal analysis and higher‑quality generated code for modern Fortran workloads.

• Easier and more scalable offload of dynamic data structures with support for the automap modifier on declare target enter for allocatables, reducing manual data‑mapping effort and helping developers write more maintainable heterogeneous Fortran applications.

Intel® oneAPI DPC++ Library

• Delivered significant performance improvements to the radix sort algorithm, enabling faster data processing workflows.
• Introduced an experimental radix sort kernel template algorithm featuring a cutting-edge SYCL implementation of the Onesweep algorithm, available in the oneapi::dpl::experimental::kt::gpu namespace for early adopters.

Intel® Distribution for GDB*

• Delivers the latest debugging capabilities and long‑term stability by rebasing to the upstream GDB 17.1 release
• Improves developer productivity and debug fidelity through expanded DWARF support, including embedded source code and new documentation for loading debug symbols for the Kobuk GFX driver, enabling more accurate, self‑contained debugging across complex toolchains and environments (DWARF Issue: DWARF and source text embedding)
• New Platform support for Clearwater Forest (CWF) and Wildcat Lake (WCL) 

Intel® VTune™ Profiler

• Added support for Wildcat Lake and Clearwater Forest processors, along with Python 3.13 and 3.14
• Improved XPU Offload analysis finalization time with the ability to select Instrumentation and Tracing Technology (ITT) top domains from both the GUI and CLI, and introduced a unified GPU / Media Hotspots analysis providing integrated performance insights across CPU, GPU, and NPU workloads
• Microsoft Visual Studio* 2019 is no longer supported; users should migrate to a supported Microsoft Visual Studio*version

Intel® oneAPI Math Kernel Library

• Improved the performance of MCG59 engine for Random Number Generation on Intel® Xeon® 6 processor with P-cores.
• Improved the performance of oneMKL Pardiso in the Conditional Numerical Reproducibility mode. It benefits HPC customers that require the results to be bitwise reproducible from run to run.
• Introduced new Intel® Advanced Vector Extensions 10 (Intel® AVX10) and Intel® Advanced Performance Extensions (Intel® APX) optimizations as a preview code-path. Use the environment variable MKL_ENABLE_INSTRUCTIONS=AVX10 or the related API to enable it. 

Intel® oneAPI Threading Building Blocks

• Delivers immediate performance gains on modern systems through internal algorithmic improvements and scheduler enhancements, including over 3× scalability gains for concurrent ordered containers on high‑core‑count platforms—allowing teams to unlock more throughput from existing hardware without changing application design.
• Simplifies optimization for complex platforms by introducing new and extended APIs that expose NUMA awareness and hybrid‑core constraints directly to the runtime, enabling oneTBB to place tasks closer to memory and on the most suitable core types for better efficiency, predictability, and scalability.
• Improves responsiveness and efficiency through new task‑level synchronization options, resource‑aware flow graph execution, and configurable runtime blocking behavior, allowing applications to reduce latency for critical work, protect shared resources, and minimize idle CPU usage on server systems.

Intel® Integrated Performance Primitives Library

• Enhanced Performance: Applications using image processing and computer vision functions will see improved performance with AVX10.2 ISA support and optimized IPP functions on AVX512 hardware.
• Large Dataset Support: Signal Processing domain now supports 64-bit array sizes, enabling processing of very large datasets that exceed previous size limitations in applications.
• Expanded API Library: 130+ new Signal Processing APIs are now available, providing additional functionality and options for implementing signal processing features in projects.

Intel® Cryptography Primitives Library

• Future-Proof Security: Applications gain protection against quantum computing threats with new ML-DSA and LMS post-quantum algorithms for key generation and digital signatures, ensuring long-term cryptographic security.
• Enhanced Cryptographic Standards: Support for PKCS#5 and new DRBG algorithms provides additional industry-standard cryptographic options, improving compliance and security implementation flexibility.
• Optimized Performance: SHA3 and ML-KEM algorithms deliver improved performance on Intell® Xeonl® 6 with P-cores, reducing cryptographic processing overhead and enhancing application responsiveness.

Intel® oneAPI Collective Communications Library

• C API set by default and functional out-of-box support for Intel® Arc™ Pro B60/B70 Graphics delivers optimized scale up/out performance, reducing setup time and improving computational efficiency for high-performance workloads.
• Intel® Implicit SPMD Program Compiler (Intel® ISPC) support for Allgather, Allreduce, Alltoall, ReduceScatter, Broadcast, and Group APIs plus enabled user-defined reduction operations accelerate distributed computing and complex multi-node processing tasks.
• Enhanced profiling tools with imbalance tracing, new memory management APIs, and SYCL graph support provide developers better visibility and control, reducing debugging time and optimizing application performance.

Intel® oneAPI Data Analytics Library

• Faster adoption of scikit-learn v1.9 features with preliminary Intel® Extension for Scikit-learn* support.
• Broader compatibility through expanded Array API standard alignment.
• Leaner installations with optional GPU components for CPU-optimized workflows.

Intel® oneAPI Deep Neural Network Library

• Optimized support for the latest Intel® processors and GPUs, delivering faster model training and inference.
• Reduced memory bottlenecks through host-side scalar memory management, optimizing critical matrix multiplication and convolution operations.
• Added support for emerging floating-point formats (MXFP8, NVFP4), enabling users to leverage state-of-the-art quantization methods that deliver faster training and inference while maintaining accuracy.

Intel® MPI Library

• MPI-5.0 standard technical preview with libfabric v2.4.0 integration, extended compatibility for Intel's latest compilers while maintaining backward compatibility, and updated IMB 2021.11 with bug fixes and improved compiler compatibility - delivering faster, more reliable parallel computing.
• Extended thread-split mode support for all communication routines and significantly improved job start latency for single-node configurations, enabling faster application startup and more efficient parallel processing workflows.
• Added support for Clearwater Forest processors and technical preview of single node Intel® Arc™ B support, providing broader hardware options and enhanced performance capabilities for high-performance computing workloads across diverse platforms.

Known Issues and Limitations

2026.1

• A versioned 2026.1 runtime container has not been published for this release due to the runtime meta-packages retain the previous version of 2026.0. Customers who require a runtime container can build one using the published Dockerfile.

Deprecations and Removals

• The Intel® DPC++ Compatibility Tool has been removed in 2026.0 release. Future contributions and issue reporting are welcomed in the SYCLomatic open-source community.

• Intel® oneAPI Deep Neural Network Library will no longer be part of the Intel® oneAPI Toolkit starting 2027.0.

System Requirements

Note: Minor releases (e.g., 2026.1) automatically inherit all OS requirements from the base major release (2026.0).Only operating systems marked with (+) are newly added in that minor release, and those marked with (–) are removed. Asterisks(**) indicates Deprecations. All others remain unchanged and are not duplicated in the table.

Hardware Requirements

Intel® software tools are optimized and validated on Intel® platforms, where full performance expectations and support are provided. While the tools may operate on other x86-based systems, those configurations are outside our validation scope and are not supported.

Supported Operating Systems

• The operating systems listed below are supported on Intel® 64 Architecture. Individual tools may support additional operating systems and architecture configurations. See the individual tool release notes for full details.
• oneCCL is not supported on Windows
• oneTBB is not supported on GPUs

Note: These OS distributions are tested by Intel or known to work; other distributions may or may not work and are not recommended. If you have questions, access the Intel Community Forums when you need assistance.  If you have Commercial Support, create a support ticket.

Development Tools

Visual Studio Code (VS Code) Extensions for Intel® oneAPI Toolkits

The VS Code extensions for oneAPI Toolkits provide assistance to developers who are creating, debugging and profiling oneAPI applications. The Using Visual Studio Code with Intel® oneAPI Toolkits User Guide provides additional detail.
The following VS Code extensions are available in the VS Code marketplace:

• Sample Browser for Intel® oneAPI Toolkits
• Environment Configurator for Intel® oneAPI Toolkits
• Analysis Configurator for Intel® oneAPI Toolkits
• GDB GPU Support for Intel® oneAPI Toolkits
• DevCloud Connector for Intel® oneAPI Toolkits

Eclipse*

• Eclipse 2025-12
• Eclipse 2025-06
• Eclipse 2026-03

Microsoft Visual Studio*

• Microsoft Visual Studio* 2022 17.x with 'Desktop development with C++' component installed
• Microsoft Visual Studio* 2026 18.x with 'Desktop development with C++' component installed

oneAPI Toolkits on Microsoft* Windows Subsystem for Linux 2 (WSL 2)

• With Microsoft* Windows Subsystem for Linux 2 (WSL2), you can use native Linux distribution of Intel® oneAPI tools and libraries on Windows*. For more information on usage, refer to Use Intel® oneAPI Toolkits on Microsoft* Windows Subsystem for Linux 2 (WSL 2)

Software Requirements

• GNU* Bash is required for local installation and for setting up the environment to use the toolkit.
• Linux kernel 4.11 or higher
• Linux Developer tools component installed, including gcc, g++ and related tools.
• The Intel® compilers are tested with a number of different Linux distributions, with different versions of gcc. The version of glibc you use must be consistent with the version of gcc in use. other distributions may or may not work and are not recommended. If you have questions and have Commercial Support, create a support ticket. Otherwise use the Intel Community Forums when you need assistance.

For GPU: Level 0 and OpenCL  graphics drivers

• To install the Linux General Purpose Intel GPUs (GPGPU) Driver, see the installation guide and follow the directions for your device.

• You must have the latest drivers to collect GPU usage events and processor graphics hardware events on Windows*. Request drivers from Intel® Support or download them here. 

• Intel® VTune™ Profiler Platform Profiler data collector requires Python* 3.6 or later
• To view PDF documents, use a PDF reader, such as Adobe Reader*

• Requirements to compile DPC++ applications: You must compile DPC++ applications with -gline-tables-only and -fdebug-info-for-profiling options to ensure the availability of debug information for performance analysis.

Graphics Driver Installation

• Windows Intel® Graphics Driver
  To install the driver follow the directions in the article appropriate for your device

  • Intel® Arc™ Graphics, 11th-13th Gen Intel® Core™ processor directions.
  • Xe Dedicated, 6th-10th Gen Intel® Core™ Processor Graphics, and related Intel Atom®, Pentium®, and Celeron® processors directions. Driver version varies depending on the Intel® Graphics in the system.
  • Intel® Data Center GPU Flex Series (ATS-M). Contact your OEM representative for access to the Intel® Registration Center.
• Linux General Purpose Intel® GPUs (GPGPU) Driver
  For all Intel® GPUs, see this article, Overview, and follow the directions for your device.

Support Deprecated

• The following OS support is deprecated and will be discontinued with a future release, See details in the System requirements section
  • For CPU
    • Debian* 11, 12
    • Fedora* 43
    • SuSE LINUX Enterprise Server* 15 SP5
    • Ubuntu LTS 22.04
  • For GPU
    • Intel® Data Center GPU Max Series(PVC): 
      • SLES15 SP4, SLES 15 SP5
    • All other client GPU platforms:
      • Ubuntu 25.10(2026.0 onwards)

Support Removed 

• The following OS is no longer supported for latest release
  • For CPU
    • RHEL 9.4
    • Rocky 9.6
    • Fedora 42
    • Debian*11
    • Red Hat Enterprise Linux* 8.10
  • For GPU
    • Intel® Data Center GPU Max Series: 
      • RHEL/Rocky 8.10, 9.5
      • SLES 15 SP4
      • Red Hat Enterprise Linux*/Rocky 10.0
    • Intel® Data Center GPU Flex Series:
      • RHEL/Rocky 8.10, 9.5
      • Windows Server 2019
      • Red Hat Enterprise Linux*/Rocky 10.0
    • All other client GPU platforms:
      • Ubuntu LTS 25.04
      • Ubuntu 25.10(2026.1 onwards)
• The following IDE is not longer supported for 2026.0

  • Support for Microsoft Visual Studio* 2019 has been discontinued starting from the 2026.0.0 release.

    Users are recommended to upgrade to a supported version of Microsoft Visual Studio* (Microsoft Visual Studio* 2022 or later) to continue using the Compiler IDE extensions.

  • Support for the DPCPP Compatibility Tool (DPCT) has been removed starting from the 2026.0.0 release.

    Users are advised to remove the DPCT extension from Microsoft Visual Studio* and Eclipse IDEs if already installed.

Other Documentation and Support

Customers with active Priority Support for Intel® oneAPI Toolkit can open a ticket for this product at Online Service Center.

• Installation Guide for Intel oneAPI Toolkits
• Get Started with the Intel® oneAPI Toolkits for Linux*
• Get Started with the Intel® oneAPI Toolkits for Windows*
• Release Notes for Intel® oneAPI Base Toolkit (Archive)
• Release Notes for Intel® oneAPI HPC Toolkit (Archive)

Release Notes for All Tools included in Intel® oneAPI Toolkit

• Intel® oneAPI DPC++ Compiler Release Notes
• Intel® oneAPI DPC++ Library Release Notes
• Intel® Fortran Compiler
• Intel® oneAPI Math Kernel Library Release Notes
• Intel® oneAPI Threading Building Blocks Library Release Notes
• Intel® Integrated Performance Primitives Release Notes
• Intel® Cryptography Primitives Library Release Notes  
• Intel® oneAPI Data Analytics Library Release Notes
• Intel® VTune™ Profiler Release Notes
• Intel® oneAPI Deep Neural Network Library Release Notes
• Intel® oneAPI Collective Communications Library Release Notes

Notices and Disclaimers

Intel technologies may require enabled hardware, software or service activation.

No product or component can be absolutely secure.

Your costs and results may vary.

© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.

No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document.

The products described may contain design defects or errors known as errata which may cause the product to deviate from published specifications. Current characterized errata are available on request.

Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.