Gemini flash actually did far better at OCRing an image than Claude and openAI I’ve found, and the price is very attractive. Going to hack on getting it integrated into a side project this weekend
Google's horrible naming in full effect. What model version is this, is their GPT4o/Claude3.5 Sonnet level model? Excited to give it a shot!
Flash = Haiku, Pro = Sonnet, Ultra = Opus
I've tried Gemini 1.5 flash, and it's about as useless (or useful, if you are a glass half full) as llama 3.1 8b running locally.
While these tokens are cheap, they're not very smart. Too many hallucinations.
I struggle to imagine a use case for this model, other than maybe generating slop.
Really? Seems to work fine for me. I have Gemini 1.5 Flash summarize articles for me and it works great.
Generally those smaller models do well when most of the data comes from the prompt. Summarising, rephrase and change tone, categorize. Ask them a question, or to generate code and you'll get hallucinations