For large models
Send ten thousand requests. Get back one job.
Instant batch inference takes a list of prompts — hundreds of them, or thousands — and runs them as a single submission. n4n fans the batch out across every available provider in parallel and works through it as fast as capacity allows: no fixed next-day queue to sit behind, and no separate batch endpoint family or pricing tier to learn beyond the one job you submit. It's built for the big, expensive frontier models, where per-request overhead — connection setup, queuing — adds up fast once you're operating at real scale.
Illustrative — example throughput for a single batch job, not a live production count.
How it works
One job in, every result out.
01
Submit the batch
Send your requests as one job — a list of prompts, or a file — instead of one call per request.
02
n4n fans it out
Your batch is distributed across every available provider in parallel, using the same routing and fallback logic as a single request.
03
Collect the results
Results come back as they finish — no fixed 24-hour turnaround to wait out.
Illustrative — batch endpoint shape, subject to change
curl https://api.n4n.ai/v1/batches \ -H "Authorization: Bearer n4n_your_key" \ -H "Content-Type: application/json" \ -d '{"model":"anthropic/claude-sonnet-5","requests":"@prompts.jsonl"}'
Which one do I want?
Batch inference vs. load factor.
Instant batch inference
For big frontier models, run at real scale. One job, parallel fan-out, no queue.
Load factor pricing
For open-source models, priced against live GPU load — up to 80% off when we're quiet.
See how it works →Start your first batch.
Sign in, create a key, and send your first batch job in under a minute.