01
One typed interface
Submit provider-native batch jobs without rewriting orchestration for every API.
Native batch jobs across seven AI providers
Ship large AI workloads through one Python API.
Submit text, embedding, and image batches through one typed async interface. Batchwork handles provider-specific serialization, polling, and normalized results at roughly 50% lower provider batch pricing.
Python 3.11+ · MIT licensed · Built by Ajanraj
batchwork.py
$ uv add batchwork
job = await client.batch(
model="openai/gpt-5.5",
requests=requests,
)
Your workload
1–50,000 requests
Batchwork
serialize · submit · poll
OpenAI
Anthropic
Gemini
Groq
Mistral
Together
xAI
Normalized results
Correlated to every request by custom_id
Install with uv
uv add batchworkInstall with pip
pip install batchworkFrom request to result
Familiar Python. Provider-native execution.
Submitting returns a BatchJob immediately. The provider processes it asynchronously—often for minutes and sometimes up to 24 hours—while Batchwork gives you one lifecycle to poll, wait, collect, or cancel.
- Models use
provider/modelsyntax. - Results correlate through
custom_id. - Provider output order never has to match input order.
quickstart.py
import asyncio
from batchwork import BatchRequest, Batchwork
async def main() -> None:
async with Batchwork() as client:
job = await client.batch(
model="openai/gpt-5.5",
requests=[
BatchRequest(custom_id="hello", prompt="Say hello")
],
)
await job.wait(timeout=3600)
for result in await job.collect():
print(result.custom_id, result.text)
asyncio.run(main())Provider coverage
Seven APIs. One batch workflow.
Text batches across every provider, plus embeddings and image generation where native batch APIs support them.
Built for real workloads
The batch plumbing is already handled.
Keep application code focused on inputs and outputs while Batchwork owns the provider-specific boundary.
02
Native serialization
Batchwork handles each provider’s request shape, including file-backed and inline submission.
03
Normalized results
Consume jobs, results, usage, and errors through consistent Python models.
04
Rich requests
Send messages, function tools, structured content, and remote media where supported.
05
Production lifecycle
Run built-in polling, persistent stores, and signed webhooks for long-lived workloads.
06
Lean dependencies
The base package uses HTTPX directly—no provider SDK or JavaScript runtime required.
Start your next batch in Python.
Install Batchwork, add one provider credential, and submit through the same async interface everywhere.