A LocalLLaMA user says Basalt ran faster and made fewer errors than Strata with the same Qwen model, after switching engines. The post describes the results as one user’s experience, not a benchmark.
The LocalLLaMA poster said they had used Strata for about a month before switching to Basalt. They reported that Strata sometimes misread its context after compaction, while Basalt seemed less error-prone in their use.
The user said the comparison used the same Qwen model, including an NVFP4 version, and reported higher output speeds on Basalt. The post does not provide a controlled benchmark or identify the hardware model, so the results reflect the poster’s setup and experience.