Yes but llama 4 It has much higher costs, it’s wasted for pure raw text extraction (for semantic extraction instead, it’s perfect). gemma 3-en4b for pure extraction is amazing and costs very little; with groq it would be lightning fast.
Good points! What are your thoughts on Gemma vs. Qwen?
(Internally we much prefer Qwen models right now)
Gemma is more efficient in terms of computational cost (not that it’s a problem for Groq given the LPU architecture). Qwen is probably also more powerful; I just hope that more multi-modal models with low-cost vision are added.
1 Like
Thanks! Yeah I think Qwen in general is more powerful… no official announcements yet but we’re working very hard on getting something like that out soon 0:)
Any updates on adding a new VLM (Qwen, Gemma or any other model better than Llama Maverick) ?
No updates or ETA yet, sorry!