# XHToken Releases Spark-X2.6-4B Bilingual Model

> A compact 4-billion-parameter instruction model targets English and Chinese tasks under a permissive Apache 2.0 license.

Published by The Open Weights on Oct 19, 2026. Canonical: https://theopenweights.com/news/spark-x2-6-4b-uzuv

## Key facts

- Company: XHToken
- Model: Spark-X2.6-4B
- Version: X2.6-4B
- Category: Text / LLM
- Modalities: Text / LLM
- License: Apache 2.0 (Open weights, commercial use allowed)
- Parameters: 4B
- Context window: 1.0M tokens
- Architecture: Spark2_5ForCausalLM
- Precision: BF16
- Weights size: 8.2 GB
- Significance: minor
- Published: 2026-10-19
- Last verified: 2026-10-28
- Hugging Face: https://huggingface.co/XHToken/Spark-X2.6-4B
- Canonical URL: https://theopenweights.com/news/spark-x2-6-4b-uzuv

XHToken has published Spark-X2.6-4B, a 4-billion-parameter instruction-tuned language model built to handle both English and Chinese. The release is available now on [Hugging Face](https://huggingface.co/XHToken/Spark-X2.6-4B) under the permissive Apache 2.0 license.

As a dense (non-mixture-of-experts) model in the 4B range, Spark-X2.6-4B sits in the increasingly crowded small-model tier, where developers trade raw capability for lower memory use, faster inference, and the ability to run on modest hardware. Models of this size are well suited to on-device deployment, fine-tuning experiments, and cost-sensitive production workloads.

## Why it matters

Bilingual coverage is the model's headline feature. A single checkpoint that handles both English and Chinese instruction-following is useful for teams building for cross-border audiences without maintaining separate models.

- **Size:** 4B parameters, dense architecture
- **Languages:** English and Chinese
- **License:** Apache 2.0, permitting commercial use and modification
- **Modality:** Text-only instruction following

The release notes on Hugging Face remain light on published benchmarks and context-length details, so prospective users will want to validate performance against their own tasks. Still, the permissive licensing and compact footprint make Spark-X2.6-4B an easy candidate for hands-on evaluation.

## Get the model

- [Hugging Face](https://huggingface.co/XHToken/Spark-X2.6-4B)

## Sources

- [XHToken/Spark-X2.6-4B](https://huggingface.co/XHToken/Spark-X2.6-4B) — Hugging Face, Oct 19, 2026

---
Source: The Open Weights (https://theopenweights.com/). Aggregated and written by Claude, curated by humans. Cite as: "XHToken Releases Spark-X2.6-4B Bilingual Model", The Open Weights, Oct 19, 2026, https://theopenweights.com/news/spark-x2-6-4b-uzuv