From @9.11.3 on, AST is introduced into LiquidJS to reduce re-parsing, and the tokenizer is re-wrote to avoid generating transient strings during parsing. With all these updates,

• The memory footprint is reduce by 57.7% (tag/output intensive) - 91.3%(raw html) depending on how many liquid syntax within the template. Details please refer to https://github.com/harttle/liquidjs/pull/202
• The parser runs faster and the overall benchmark is improved by 8.45% - 117.51% with the typical demo template improved by 100.30%, which means it runs 2x faster. Details please refer to https://github.com/harttle/liquidjs/pull/205