SparkLLM releases Spark-X2.6-4B, served day 0 by vLLM

SparkLLM releases Spark-X2.6-4B, a 4-billion-parameter model with a 1M-token context window under Apache 2.0.

The model supports controllable thinking at none, low or high levels. vLLM, the open-source inference engine, says it serves the model on day 0, the day of release. SparkLLM's announcement, as quoted in the report, says the model adds stronger coding, agent and long-context skills through improved post-training.

The report comes from vLLM's X account and repeats the figures SparkLLM gave. It does not include independent benchmark results.