A Reddit user has expressed concern regarding recent changes made by Google to its Gemini models, particularly regarding increased costs and decreased quality.

Increased Costs, Decreased Quality

The user, who used Gemini 2.0 Flash for OCR (Optical Character Recognition), data extraction, and summarization tasks, noted an approximately sixfold increase in output costs with subsequent versions (Gemini 2.5/3). At the same time, accuracy did not improve and, in some fine-tuning tests, even worsened.

End of the Economical Model

The cheapest option, Gemini 2.5-flash-lite, has been marked as EOL (End Of Life) without a 3-flash-lite successor being announced. This makes long-term planning difficult for those who rely on these models for specific tasks.

Search for Alternatives

The user has opened a ticket with Google to request an EOL extension for Gemini Flash models or a low-cost successor. In the meantime, they are looking for an alternative LLM for OCR and data extraction that offers simple and managed fine-tuning at a reasonable price.

For those considering on-premise deployments, there are trade-offs to consider. AI-RADAR offers analytical frameworks on /llm-onpremise to evaluate these aspects.