Join us live to see what developers can build with NexaSDK on @Snapdragon—plus a live demo and open Q&A with the team. 🚀

What an incredible 2025! Thankful for the partnerships, trust, and collaboration that made it possible. Excited for what we’ll build together next 🚀

We just launched NexaSDK for Mobile on Product Hunt. Try out NexaSDK for Mobile and we’d love your feedback: producthunt.com/products/nex…

Huge appreciation to our partners from @Microsoft @GoogleDeepMind @Qualcomm @NVIDIA @IBM @AMD @Intel @Qwen and so many others who featured us on stages, blogs, and launches!

gpt-oss-20b running on Hexagon NPU via Nexa SDK 🔥

Thanks @nvidia @NVIDIA_AI_PC for promoting our Hyperlink product!

Following the launch of the Nexa Android SDK, we ran a 10-minute LLM stress test on the Samsung S25 Ultra with Qualcomm Hexagon NPU: ⚙️ CPU: throttled from ~37 t/s → ~19 t/s at 42 °C ⚙️ NPU (Qualcomm Hexagon): held steady at ~90 t/s and 36–38 °C — 2–4× faster under load 🔋 Both used ~6% battery, but the NPU delivered ~3.7× more tokens per battery Efficient, cool, and fully offline — that’s the power of NexaSDK on-device inference. 🔥📱 📘 Learn more: docs.nexa.ai/nexa-sdk-androi…

Today, we’re launching Android Java & Kotlin support for NexaSDK (Beta) — bringing the full power of on-device AI to billions @Android phones powered by @Qualcomm @Snapdragon chipsets. This is a major leap forward for the world’s largest mobile developer community: ✅ Seamless NPU / GPU / CPU inference with a single SDK ✅ Run top-tier open models from @Google , @nvidia , and more ✅ 100% private, zero cloud latency, and fully offline

Our work on NPU-accelerated inference for SDXL-Turbo has been featured by AMD on their official blog. We also have been invited to the PyTorch Conference with AMD. Try it out on your AMD laptop: nexa infer NexaAI/sdxl-turbo-amd-npu

LFM2-1.2B models from Liquid AI are now running fully accelerated on Qualcomm NPUs via the NexaML engine — real-time performance with minimal memory use, right on the edge. Four new variants power everything from chat to document parsing: 💬 LFM2-1.2B – general chat & reasoning 🔍 LFM2-1.2B-RAG – retrieval-augmented local chat ⚙️ LFM2-1.2B-Tool – structured tool use & agent workflows 📄 LFM2-1.2B-Extract – ultra-fast document parsing A hybrid Transformer SSM design few frameworks can even run — but NexaML can. Already seeing what this unlocks for partners like Brilliant Labs in AR/VR. ⭐ Star NexaSDK to catch new edge model drops : github.com/NexaAI/nexa-sdk

🚀 we’ve achieved Day-0 full-platform inference support for Qwen3-VL-4B! From NPU to GPU to CPU, across Qualcomm, Apple, AMD, Intel, MediaTek, and NVIDIA, you can now run the latest multimodal model locally, natively, and at full speed — all powered by NexaSDK. This breakthrough comes from building our core inference engine, NexaML, from the ground up — deeply integrating GGML and MLX, and optimizing for every hardware architecture. In close collaboration with @Alibaba_Qwen , and partners at @Qualcomm , @MediaTek , @AMD, and @intel, we’ve delivered a truly Unified Inference Engine — bridging the compute gap between devices. 🌍 The future of multimodal AI is here — not in the cloud, but in your hands. One command. All platforms. Full performance.

We have provided day-0 support to run Qwen3-VL on NPU / GPU / CPU, try it here: huggingface.co/collections/N…

Thanks @simonw for mentioning our work! We continue to compress and prune gpt-oss such that it can fit in latest iPhone. More exciting updates to come soon!

Thrilled to share that we just launched another Day 0 support: Qwen3-VL-30B-A3B-Instruct is now live on NexaSDK with full Apple Silicon GPU support. You can try it right now with: nexa infer NexaAI/qwen3vl-30B-A3B-mlx Thanks @JustinLin610 for partnership and @awnihannun for the MLX work!

🚀 Granite-4.0 day-0 support with NexaML According to our partners, this is the first-ever NPU day-0 support for a new LLM — Granite 4.0 is live on Qualcomm NPUs with NexaSDK. With NexaSDK, you can run Granite 4.0 seamlessly across NPU, GPU, and CPU — and switch between them with a single line of code. On AMD / Intel / Qualcomm / Apple GPU → nexa infer NexaAI/granite-4.0-micro-GGUF On Qualcomm NPU → nexa infer NexaAI/Granite-4-Micro-NPU This unlocks Granite-4.0 for PCs, smartphones, automotive, and IoT — ready from day zero. 💡

Running AI models locally on the @Snapdragon X Elite just went to a WHOLE NEW LEVEL thanks to @nexa_ai! I can't wait to be able to test this on the X2 Elite Extreme which will be up to 78% faster. Check this out on the @Surface Laptop

NexaSDK brings Intel NPU, GPU, and CPU into one stack for real-time, private, local AI. Star us: github.com/NexaAI/nexa-sdk Try it: sdk.nexa.ai/model/Llama3.2-3…

Now with CPU, GPU, and Snapdragon NPU support in one unified architecture—packed into a lightweight 60MB installer. No more juggling installers, APIs, or backend-specific builds. ⭐ If Nexa SDK helps you, give us a star: GitHub: github.com/NexaAI/nexa-sdk Blog: nexa.ai/blogs/backend