Paid inference / launch workload review
Prove the workload before you commit.
National Compute is qualifying DeepSeek V4 Flash 0731, GLM 5.2 and MiniMax H3 behind an OpenAI-compatible API. Bring the workload you already run. We will measure model fit, latency, concurrency and cost before discussing a longer commitment.
The initial review is a technical fit check, not a service commitment.
01 / launch models- 01DeepSeek V4 Flash 0731
- 02GLM 5.2
- 03MiniMax H3
OpenAI-compatible access. Limits and pricing publish per model after qualification.
A smaller first decision
From review to recurring service.
The first conversation should establish fit, not force a contract. Each later step has a price, boundary and decision criterion before it starts.
- 01
Review the workload
A 30-minute working session covers your current model, traffic, spend, data constraints and the result that would justify a switch.
No purchase commitment - 02
Measure a real envelope
If the service fits, we scope a bounded evaluation against agreed quality, latency, concurrency and cost criteria.
Bounded evaluation - 03
Commit after evidence
Only a passing evaluation moves to a fixed-rate 90-day production service with a defined operating boundary and support owner.
Recurring service
Workload filter
One service should solve one clear problem.
A workload that needs a different product is a useful disqualification, not a reason to expand the first launch.
Good first workload
- OpenAI-compatible text chat completions
- Recurring usage with a known traffic or spend envelope
- Synthetic or approved evaluation data
- A technical owner who can judge response quality
Not supported at launch
- Regulated data or unqualified residency requirements
- Custom training, fine-tuning or bespoke model work
- Streaming, tools, image, audio or embeddings
- Dedicated capacity before a measured evaluation
Start with five facts
Bring the workload, not a buying decision.
Send the workload, current provider and model, monthly usage or spend, expected concurrency and data constraints. We will reply with fit, disqualification or the next measurement.