Reddit user reports Basalt outperforming Strata in local AI tests
AI summary
A Reddit user compared Strata CLI engine 0.1.59 with Basalt 0.3.2 on Windows using Qwen3.8-Flash-Next IQ3S, a 256k context, and vision enabled. The user reported average generation speeds of 99.2 tokens per second for Strata and 100–118 tokens per second for Basalt, while Basalt also used less RAM and produced smoother, more reliable task completion. Basalt's peak prefill speed was reported as 4,200, although the user's average figure was uncertain.
In-depth insight
The single-user test provides a practical data point for people choosing between two local-model runtimes, but its reliability findings remain subjective and unverified.