Just got the email blast and dug into the docs. Humata has officially rolled out their API v2, and the billing model has been completely overhauled. This isn't a minor tweak—it's a fundamental shift from a simple per-document/page model to a consumption-based system centered on "Tokens." If you're building any automated pipelines against their API, this will break your cost projections and likely your current integration code.
The old model was straightforward, if a bit crude: you paid X per document or per X pages uploaded. Predictable, but not necessarily fair for smaller files. The new model is all about input and output tokens, much like the LLM APIs we're all used to (OpenAI, Anthropic, etc.). They've even provided a token estimator in their dashboard.
Here’s the critical breakdown of the new units:
* **Upload Tokens:** Consumed when you **add** a file to Humata. This is based on the raw text content extracted. Their estimator suggests a 100-page PDF is roughly 50K tokens.
* **Question Tokens:** Consumed when you **query** your data. This includes the tokens from the context retrieved *plus* your question *plus* the generated answer. A complex query pulling from multiple documents will burn through these fast.
* **Tokens are pooled** across both types from your monthly limit.
The immediate implications for pipeline developers:
1. **Cost unpredictability:** A pipeline that processes 1000 PDFs one month and queries them heavily will cost orders of magnitude more than a month where you just upload and archive. Your burn rate is now directly tied to user query activity.
2. **Monitoring imperative:** You **must** now instrument your integration to track token consumption per operation, similar to how you'd monitor LLM API costs. Blindly firing queries will get expensive.
3. **API Changes:** The v1 endpoints are deprecated. The new `/v2` endpoints have different request/response schemas. The `question` endpoint, for example, now returns a breakdown of token usage in the response body.
Example of the new API response snippet you need to handle:
```json
{
"answer": "The report indicates a 15% increase...",
"sources": [...],
"usage": {
"upload_tokens": 0,
"question_tokens": 1247,
"total_tokens": 1247
}
}
```
My early take: This move aligns Humata with the standard paradigm for GenAI services, which is logical for them but a headache for anyone on fixed budgets. It makes small-scale, automated processing cheaper, but large-scale Q&A systems potentially much more expensive. You can no longer just budget for ingestion; you have to budget for *conversation*.
If you're using this in production, you need to audit your integration this week. Update your client code to use the v2 endpoints, implement usage logging, and re-forecast your monthly spend based on actual query patterns. The old pricing is gone.
fix your schema
fix your schema