OpenAI Flex Processing

[compatible9]

Flex processing provides significantly lower costs for Chat Completions requests in exchange for slower response times. It is ideal for lower-priority tasks such as parsing documents with OpenAI

Please considere this implementation, more information at https://platform.openai.com/docs/guides/flex-processing?api-mode=chat

[ZhenhangTung]

Thanks for the input. Actually most of our developers complain about low speed of parsing documents, so I'll think about it and see if there is any suitable use cases for this feature. And since we're a platform, I'll also take a look at other LLMs to see if they have similar feature.

[compatible9]

I my case I parsed about 100k documents some of them with 80 and up +100 pages each, as you can see using raptor is a must, I could save a few houndrees of dollar using flex or even batch processing (https://platform.openai.com/docs/guides/batch)

[onestardao]

interesting to hear this is a recurring pain point — we’ve seen similar frustration from teams trying to parse PDFs, scanned documents, or hybrid-text formats using OpenAI endpoints.

the issue often isn’t just speed, but semantic loss during parsing, especially when layout-dependent meaning is involved (e.g., tables, footnotes, nested lists).

we ended up building a fault-tolerant pipeline for semantic parsing under lossy/fragmented inputs — if you’re curious, I can share the structure or failure maps we used internally. might help pressure-test flex mode too.

just let me know if it's useful.

[ZhenhangTung]

@onestardao Hi, thanks for the input. I'm the product manager. Would you love to join the community and build RAGFlow to make it better together? :)

[onestardao]

thanks for the invite really appreciate it. i’m slammed with other commitments right now so i can’t join the community. happy to share the ProblemMap checklist or give quick feedback on specific issues if you ping me here. best of luck building this out.

^____^ BigBig