Hello,
I’m planning a one-time, large-scale OCR project using the Gemini API and would appreciate guidance before I set anything up. I haven’t attached a billing account yet, so no charges have been incurred —
I’d like to plan the approach correctly first.
Project details:
- Task: OCR text extraction from 316,000 photographed document pages, each photo/page at 1400x800 DPI.
- Total dataset size: approximately 46 GB
- Planned model: Gemini 3.5 Flash
- Processing method: asynchronous Batch API, split into multiple input files (up to the 2 GB file cap) and uploaded in waves due to the 20 GB total batch storage limit
- Could you please estimate the cost of such a task at current published Batch API pricing, for the full dataset?
My questions:
On a new billing account (Tier 1), the enqueued-token cap for Gemini 3.5 Flash is listed as 3,000,000 tokens, versus 400,000,000 at Tier 2. Since this is a one-time, clearly bounded job, is it possible to request an expedited or temporary rate limit increase ahead of naturally qualifying for Tier 2 (which normally requires $100 spend + 3 days)? I understand approval isn’t guaranteed, but I’d like to submit a request if that’s the right channel.
2. Are there any additional considerations specific to Batch API usage for Gemini 3.5 Flash I should know before committing a job this size to it?
3. Are there known best practices for structuring a 20+ file batch upload sequence (upload, process, delete, repeat) to avoid hitting the 20 GB storage cap mid-job?
4. Is there a way to track or confirm tier-upgrade status in real time once I do cross the $100/3-day threshold, so I can time my batch submissions efficiently?
5. What is the estimated time for completing such a task, in asynchronous Batch processing mode?
Best regards,
Bar