Voice AISample
Gemma 4B on H100s
- Workload
- Voice agent, live calls
- Optimise for
- Latency
- TTFT p95
- 200 ms
- Peak load
- 40 requests a second
- Uptime
- 99.9%
- Region
- US
- Test
- Shadow workload on a copy of live calls
- Optimisation
- FP8 allowed, under 1% quality loss
For teams that buy GPUs and inference
Post your workload. Get offers. Compare L1, L2 and the best quality vendor side by side.
Backed by![]()
GPU prices
The average on-demand list price across providers. We track the same providers every day, so one new name can't move the line.
List prices, not what people paid.
See every price, day by dayproviders we read prices from. Your request goes to the ones that fit.
Vendors we read prices from
How it works
Vendors that can run your model at your traffic and latency.
Every offer on the same spec. The two cheapest side by side, and the best on quality.
Quotes, follow ups, terms. You make the call.
Set your terms
Voice AISample
Document intelligenceSample
Coding agentSample
Samples. On the board, buyers' names and companies are never shown.