CreativeBaker87Q: Question about models - What to use?
David, you’ve mentioned this a few times, mostly on YouTube but maybe also here. It would be really helpful—both for me and others—if you could share some tips on the best practices for using Models in SheetXAI. I know this evolves, but the goal is to understand which models are sufficient for different use cases. Starting with the "default model" and then covering recommended options for tasks like image or video generation would be great. I know, many have their own preferences, but your insights would add valuable guidance. Thanks!
David_SheetXAI
Dec 31, 2025A: Hey CreativeBaker87 👋!
Great question! This definitely evolves over time, but let me break down the best practices for using models in SheetXAI based on what's working best right now.
FIRST - IF YOU GO THE BYOK ROUTE:
Use OpenRouter for all text stuff including the chatbot, and Fal AI for all video, image, and audio generation. That combo gives you the best coverage and pricing.
CHATBOT TAB MODELS (FOR TEXT OPERATIONS):
GPT-4o mini - This is the cheapest and best. Honestly, can't get much better than that. It can do 90% of things you throw at it. Start here for basically everything.
Claude Haiku - Use this only if you try something with 4o mini that fails TWICE. Remember that when something fails, I typically recommend you clear chat and ask again. But if it still doesn't work after a second try, switch to Haiku.
Claude Sonnet - This one is pretty expensive, but it can be used with the most complex cases. For example, if I'm building a dashboard, Sonnet is the one - it's just too powerful. Use this when you need absolute top-tier performance for complex operations.
Those are your top 3 in the chatbot tab that you should select based on complexity and budget.
SAI CONTENT TAB (FOR CONTENT GENERATION):
For most things I would recommend any of the Gemini Flash models. I believe the latest full one is Gemini 2.5 Flash. You can upgrade to Claude Haiku or Sonnet here too since they're very good at human-like speaking, but once again price-wise, Gemini Flash can't be beat for how much it can bring at such a low price.
SAI IMAGE ANALYSIS TAB:
Just copy whatever you're using in the Content tab. Same models work great here.
SAI URL TAB (FOR WEB SCRAPING/BROWSING):
Perplexity Sonar is the best - once again can't be beat. I can't even genuinely give you any competitors that can do it better. It has built-in live internet search capabilities that make it perfect for this.
IMAGE GENERATION:
If you use Fal AI, I recommend the Minimax Hailuo latest model. But this one might require testing different models to get a feel for what style and quality you prefer. Image generation is pretty subjective, so experiment a bit.
VIDEO GENERATION:
Best right now is Sora 2, but I still think any of the Minimax models can make something tangible for cheap if budget is a concern.
AUDIO GENERATION:
ElevenLabs models are the king here. No contest for voice quality and naturalness.
QUICK SUMMARY:
Default starting point: GPT-4o mini for text, Gemini Flash for content generation, Perplexity Sonar for URLs, ElevenLabs for audio.
When you need more power: Claude Haiku is your mid-tier upgrade, Claude Sonnet is your nuclear option for complex stuff.
For media: Fal AI provider with Minimax/Sora models depending on your budget.
Hope that helps clarify things! This is current as of right now, but yeah, it evolves as new models come out.
Let me know if you have other questions! 🙂