SparkLLM open-sources X2.5 4B/1.7B on-device models with native 1M-token context

vllm_project · x · 2026-09-01

SparkLLM has open-sourced Spark-X2.5-4B and X2.5-1.7B, compact on-device agent models claimed to be the only ones with native context up to 1,000,000 tokens. Built on a hybrid-attention architecture and tuned for agents, code, math and instruction following, they support 200+ languages. The vLLM project confirmed day-0 support for both via Spark's out-of-tree general plugin, so they can be served with vLLM from launch day.

Related event: SparkLLM Open-Sources X2.5 On-Device Models with 1M Context(2 posts)→

Original post →

More from coding & agent

coding & agent channel →