Jobs, not infrastructure
Post a job and walk away. We queue it, retry the rows that fail and write results back.
Millions of promptsrun asynchronously
Send a million prompts and collect the results when they land. Up to 50% off serverless rates, with no queue to babysit and no infra to size.
Up to 1M requests per job
Batch job · job_8f21cCompleted
Requests processed1,000,000 / 1,000,000
ModelKimi K3
Submitted03:40 UTC
Completed10:00 UTC · 6h 20m
Failed rows0
Batch rate applied₹2,28,000.00 · saved 50%
- Requests per job
- 1M
- Turnaround
- 24 h
- Serverless rate
- up to 50%
What you get
A queue that never becomes your problem.
Same models, same answers
A batch result matches what production will return.
Failure handling built in
Per-row error reporting with exponential retries. One malformed line never sinks a million-row job.
Deciding
Half price, at the cost of waiting.
Stay serverless when
A person is waiting
- You are driving an agent loop or a chat interface
- Tokens need to stream onto a screen as they arrive
- Traffic is interactive and hard to predict
- A 24-hour turnaround would break the experience
Use batch when
Nobody is watching the clock
- You are scoring, classifying or enriching a dataset offline
- You are running evals across many model variants
- You are generating synthetic training data
- Latency is measured in hours, not milliseconds