XHToken releases Spark-X2.6-4B, a 4B on-device language model
XHToken posted Spark-X2.6-4B on Hugging Face, a compact 4B‑parameter model designed for on‑device deployment. It builds on the Spark‑X2.5-4B architecture and gains capability through improved post‑training.
Key points
- Spark-X2.6-4B is a 4B‑parameter on‑device LLM released by XHToken on Hugging Face.
- It claims state‑of‑the‑art performance among 4B models and matches 9B models on several benchmarks.
- The model supports up to 1,048,576‑token context and three thinking levels (none, low, high) via the provided Docker images.
The release notes stronger coding, agentic, long‑context, multilingual and reasoning abilities, claiming state‑of‑the‑art performance among similarly sized open‑source models and matching or surpassing models with approximately 9B parameters on several benchmarks. The model offers three configurable thinking levels—none, low, high—and supports a maximum context length of 1,048,576 tokens. The repository provides installation via transformers==4.57.1, example code with maxnewtokens 131072, and Docker images such as lmsysorg/sglang:nightly-dev-cu13-20260827-20621aa1 and Ascend images for NPUs. Additional support for Apple silicon, Linux CPU and NVIDIA CUDA is available through Spark‑MLX‑LLM, with optional use of Ollama (v0.34.1+) or LM Studio (0.4.0+). The model series is released under the Apache 2.0 License.
XHToken/Spark-X2.6-4B · Hugging Face
huggingface.co · 20 October 2026
Loading the full article…
This text was published by huggingface.co. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
1source- Reddit discussionreddit.com
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- The Truth Behind 'Apple and Meta Team Up on AI': It's Not Siri, It's Mac Studio | A Manager's Breakdown of Confirmed Facts, Reports, Unannounced Details, and Im · 2 src
- Elevenlabs launches two realtime speech-to-speech models for AI conversations · 1 src
- Nvidia releases Cosmos3-H-Dreams real-time generative surgical simulator · 2 src
- OpenAI improves code migration for COBOL with GPT-6.1 Sol achieving 97.9% pass rate · 5 src
- Amazon fine‑tunes GPT‑OSS‑20B on SageMaker HyperPod to reach 100% multilingual reasoning · 1 src
Comments
via GitHub Discussions