Google: Gemma 4 26B A4B
DeepInfra is the cheapest and Scaleway the fastest. Novita costs more for less speed than a provider on the line.
- Live providers
- 3
- Same model, one API
- Cheapest
- $0.1375 blended
- DeepInfra
- Fastest output
- 232 tok/s
- Scaleway
- Quickest first token
- 0.23 s
- Scaleway
- Best uptime
- 100%
- DeepInfra
- DeepInfra serves Google: Gemma 4 26B A4B cheapest, at $0.1375 per million tokens blended.
- Scaleway serves Google: Gemma 4 26B A4B fastest, at 232 tok/s (median over 24 hours).
- Scaleway starts answering Google: Gemma 4 26B A4B quickest, first token in 0.23 s (median over 24 hours).
- DeepInfra has the best uptime for Google: Gemma 4 26B A4B, 100% over 72 hours.
Scaleway232 tok/s
DeepInfra93 tok/s
Novita78 tok/s
Scaleway0.23 s
DeepInfra0.69 s
Novita0.69 s
DeepInfra$0.1375
Novita$0.1975
Scaleway$0.3601
3 providers
0100200300$0.00$0.10$0.20$0.30$0.40$0.50
DeepInfra$0.1375 blended, 93 tok/s
Scaleway$0.3601 blended, 232 tok/s
Novita$0.1975 blended, 78 tok/s
Blended price, $ per million tokens (3 input to 1 output)DeepInfra$0.1375 blended, 93 tok/sScaleway$0.3601 blended, 232 tok/sNovita$0.1975 blended, 78 tok/sCustomer traffic on this model is still below the publishing floor, so its figures are from llm.buzz's test requests, tagged "test".
The last 24 hours, by hour
Last 24 hours, from llm.buzz's test requests
- DeepInfra93 tok/s
- Scaleway232 tok/s
- Novita78 tok/s
Data table
| Provider | From (UTC) | p50 | p90 | p99 |
|---|---|---|---|---|
| DeepInfra | Oct 26, 07:00 | 93 tok/s | 93 tok/s | 93 tok/s |
| DeepInfra | Oct 26, 12:00 | 15 tok/s | 15 tok/s | 15 tok/s |
| DeepInfra | Oct 26, 16:00 | 134 tok/s | 134 tok/s | 134 tok/s |
| DeepInfra | Oct 26, 21:00 | 134 tok/s | 134 tok/s | 134 tok/s |
| DeepInfra | Oct 27, 01:00 | 65 tok/s | 65 tok/s | 65 tok/s |
| DeepInfra | Last 24 hours | 93 tok/s | 15 tok/s | 15 tok/s |
| Scaleway | Oct 26, 05:00 | 232 tok/s | 232 tok/s | 232 tok/s |
| Scaleway | Oct 26, 10:00 | 232 tok/s | 232 tok/s | 232 tok/s |
| Scaleway | Oct 26, 14:00 | 161 tok/s | 161 tok/s | 161 tok/s |
| Scaleway | Oct 26, 19:00 | 161 tok/s | 161 tok/s | 161 tok/s |
| Scaleway | Oct 26, 23:00 | 232 tok/s | 232 tok/s | 232 tok/s |
| Scaleway | Oct 27, 03:00 | 161 tok/s | 161 tok/s | 161 tok/s |
| Scaleway | Last 24 hours | 232 tok/s | 161 tok/s | 161 tok/s |
| Novita | Oct 26, 07:00 | 26 tok/s | 26 tok/s | 26 tok/s |
| Novita | Oct 26, 11:00 | 193 tok/s | 193 tok/s | 193 tok/s |
| Novita | Oct 26, 15:00 | 112 tok/s | 112 tok/s | 112 tok/s |
| Novita | Oct 26, 20:00 | 37 tok/s | 37 tok/s | 37 tok/s |
| Novita | Oct 27, 00:00 | 54 tok/s | 54 tok/s | 54 tok/s |
| Novita | Oct 27, 04:00 | 78 tok/s | 78 tok/s | 78 tok/s |
| Novita | Last 24 hours | 78 tok/s | 26 tok/s | 26 tok/s |
- DeepInfra0.69 s
- Scaleway0.23 s
- Novita0.69 s
Data table
| Provider | From (UTC) | p50 | p90 | p99 |
|---|---|---|---|---|
| DeepInfra | Oct 26, 07:00 | 0.69 s | 0.69 s | 0.69 s |
| DeepInfra | Oct 26, 12:00 | 0.83 s | 0.83 s | 0.83 s |
| DeepInfra | Oct 26, 16:00 | 0.58 s | 0.58 s | 0.58 s |
| DeepInfra | Oct 26, 21:00 | 0.69 s | 0.69 s | 0.69 s |
| DeepInfra | Oct 27, 01:00 | 1.00 s | 1.00 s | 1.00 s |
| DeepInfra | Last 24 hours | 0.69 s | 1.00 s | 1.00 s |
| Scaleway | Oct 26, 05:00 | 0.23 s | 0.23 s | 0.23 s |
| Scaleway | Oct 26, 10:00 | 0.28 s | 0.28 s | 0.28 s |
| Scaleway | Oct 26, 14:00 | 0.23 s | 0.23 s | 0.23 s |
| Scaleway | Oct 26, 19:00 | 0.28 s | 0.28 s | 0.28 s |
| Scaleway | Oct 26, 23:00 | 0.23 s | 0.23 s | 0.23 s |
| Scaleway | Oct 27, 03:00 | 0.23 s | 0.23 s | 0.23 s |
| Scaleway | Last 24 hours | 0.23 s | 0.28 s | 0.28 s |
| Novita | Oct 26, 07:00 | 0.83 s | 0.83 s | 0.83 s |
| Novita | Oct 26, 11:00 | 1.00 s | 1.00 s | 1.00 s |
| Novita | Oct 26, 15:00 | 0.58 s | 0.58 s | 0.58 s |
| Novita | Oct 26, 20:00 | 0.83 s | 0.83 s | 0.83 s |
| Novita | Oct 27, 00:00 | 0.69 s | 0.69 s | 0.69 s |
| Novita | Oct 27, 04:00 | 0.48 s | 0.48 s | 0.48 s |
| Novita | Last 24 hours | 0.69 s | 1.00 s | 1.00 s |
Uptime
Last 72 hours, by hourThis model
DeepInfra
Scaleway
Novita
- Up: 95% or more served
- Degraded: 85% to 95%
- Down: below 85%
- No data
- From llm.buzz's test requests
How we measure
- Speed and first tokenOutput speed and time to first token are medians over the last 24 hours, per provider, from text requests.
- UptimeThe share of attempts the provider served over the 72 hours before the figures' end, provider failures counted against it.
- Two sourcesCustomer traffic once enough organizations count toward a figure; below that, llm.buzz's test requests, short streamed requests sent to every live provider. The tables tag those "test".
- PriceEach provider's price per million tokens, with a running discount applied. The blended price weighs input and output 3 to 1, a chat request's usual mix.
Let the router pickName the model; llm.buzz sends each request to a provider and moves to the next when one fails.