I built batch-ai, a TypeScript SDK that simplifies batch processing across AI model providers (OpenAI, Anthropic, and soon Google Gemini & xAI Grok).

Why?

If you've ever tried running AI workloads in batch mode, you know how frustrating it is:

OpenAI requires file uploads for batch processing.

Anthropic doesn't, but has different request/response formats.

Cost savings are significant (~50% vs. real-time API calls), but every provider handles batching differently.

I wanted a simple way to process large volumes of AI requests across different providers without dealing with writing code for all of them. batch-ai provides a single interface to handle batch requests efficiently.

Features:

Unified batch processing API for OpenAI & Anthropic (more coming).

Define output schemas with Zod for structured responses.

Reduce costs by using batch APIs instead of real-time calls.

Easily switch providers without changing request logic.

Who is this for?

AI moderation tools (like Filtyr, my AI content moderation SaaS). https://filtyr.com

Large-scale AI processing (e.g., sentiment analysis, classification).

Researchers & enterprises handling structured AI output at scale.

What’s next?

Support for Google Gemini & xAI Grok.

More batch API options (e.g., generateTextBatch).

Smarter retries & error handling.

Repo: GitHub – grantsingleton/batch-ai

Would love feedback from anyone working with batch AI APIs. What’s your experience? What pain points have you run into?