GitHub radar
Local LLM with 1M context that fits any laptop
XHToken (SparkLLM Team) released Spark-X2.5-4B: a compact 4B model with a 1 million token context window, 200+ language support, and GGUF versions for Ollama and LM Studio.
Spark-X2.5-4B is a compact general-purpose language model from XHToken (SparkLLM Team) with a 1 million token context window — exceptionally large for a model of its size. It supports over 200 languages including Russian and handles conversation, writing, translation, reasoning, coding, and agentic workflows. The 4.1B parameter model fits on any laptop: the 4-bit compressed version weighs around 2.5 GB and runs from 8 GB of RAM without a GPU. Fifteen quantized variants are available for llama.cpp, Ollama, LM Studio, MLX, and vLLM. XHToken built it on a custom hybrid attention architecture that efficiently processes very long inputs.
Why a vibe-coder should care
A 1M token context window means you can load an entire book, a long contract, or a large codebase and have a proper conversation without the model losing the thread. At 4B parameters it runs on ordinary laptop hardware with no GPU. That combination of lightweight size and large memory makes it stand out from other small local models.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install the Spark-X2.5-4B language model locally from https://huggingface.co/XHToken/Spark-X2.5-4B — pick the right GGUF quantization for my laptop, set it up via Ollama or LM Studio, and show me how to use it.
A regular laptop with 8 GB of RAM or more — the 4-bit version takes around 2.5 GB, no GPU needed.
Open on Hugging Face▌ More finds