Glad to hear quality comes first! Then I assume you have some public benchmarks like the ones you mention that are reproducible? I could only find this graph https://docs.morphllm.com/guides/apply but there is no mention of what it refers to, what data it used etc.