Testing 7 Free LLMs for Formatting Recipes

Amarsir · reddit · 2026-07-15

The author fed the same messy recipe text to 7 free chat models, asking them to format it into a better Markdown document, and compared their usability. Participants included Gemini Flash Extended, Deepseek Instant/Deepthink, Muse Spark 1.1, GLM 5.2, MiMo 2.5 Pro, Grok 4.0 Fast, and Kimi 2.6 Thinking. GPT and Sonnet were only used for grading.

Main Observations

Core Conclusion

The author's evaluation criteria were: retain all information, avoid unauthorized rewrites, and reliably output copyable Markdown. Based on the results, Deepseek excelled at information retention and structuring, despite some accidental deletions. Gemini performed the worst in terms of usability and fidelity.

Original post →

More from Models

Models channel →