AI developer tools · What shipped

DeepSeek and Huawei released open-source programming tools for Ascend AI chips, including computation and communication libraries plus TileLang support, designed to reduce dependence on Nvidia's ecosystem. The tools aim to simplify programming while enabling developers to fully leverage hardware performance, with optimization work completed on a supernode system built around 128 Ascend 950 chips to address efficient calculation and fast data movement between accelerators.

Do Inference Now (DIN) Deploy is an open-source collection of C++ samples that enables developers to build hardware-accelerated local AI applications. It combines ONNX Runtime with NVIDIA TensorRT RTX execution provider, supporting deployment on Windows and Linux with x86-64 and Arm64 architectures. The toolkit includes Python exporters for converting model checkpoints from Hugging Face to ONNX format, plus native C++ CLI implementations that keep model conversion separate from deployment logic. DIN Deploy provides samples for automatic speech recognition (supporting OpenAI Whisper, NVIDIA Parakeet TDT, and NVIDIA Nemotron ASR Streaming), image and video segmentation with Meta SAM 2.1, and image generation with FLUX.2-klein-4B. Performance benchmarks show significant GPU acceleration: Whisper-large-v3-turbo achieves 58.5x speedup on DGX Spark GPU versus CPU, while SAM 2.1 reaches 38.3 FPS on GPU compared to 0.5 FPS on CPU. The samples demonstrate post-training quantization via NVIDIA Model Optimizer, graphics API interoperability with Vulkan and DirectX, and tensor data management through ONNX Runtime's copy tensor API.

Overmind announced the open source release of its platform for building, fine-tuning and deploying specialized small language models. The platform provides an end-to-end workflow covering agent observability, data preparation, training, evaluation, fine-tuning and deployment, allowing teams to benchmark custom models against real-world tasks. Early deployments showed significant results including a 96% reduction in invoice-processing costs in one fintech deployment and an 86% accuracy improvement compared to frontier models in a legal benchmark. The company, founded in 2025 by former British Intelligence and fintech leaders, has recorded tens of thousands of SDK downloads and is working with organizations across fintech, legal, healthcare and cybersecurity. Security features include SOC 2 and ISO compliance.

Microsoft released updates across its AI developer tooling and infrastructure: OpenAPI 3.2.0 now supports HTTP QUERY operations, Server-Sent Events streaming, and improved binary/file schemas for modern API patterns; JSON Schema 2020-12 adds recursive schema support without brittle $ref chains; ASP.NET Core in .NET 11 adds OpenAPI 3.2.0 as first-class web API experience; Visual Studio Code now supports JSON Schema 2020-12 with built-in OpenAPI validation; Azure API Management deployed OpenAPI 3.2.0 support across regions; TypeSpec improved its OpenAPI-to-TypeSpec conversion tool for production lift-and-shift scenarios. These specifications are positioned as foundational infrastructure for AI-assisted development, improving reliability of LLM-generated code through precise API descriptions.