LLM's specifically finetuned to be reasoning engines. Higher experimental versions should usually indicate higher quality models.