DeepSeek V4 Pro repriced across 5 hosts: Relace input down 57%, Ionstream input up 169%
Published 29 September 2026
Price move, per 1M tokens
- Ionstream · input↑ 169%$0.226 → $0.608
- Ionstream · outputunchanged$1.96 → $1.96
- Ionstream · cache readunchanged$0.088 → $0.088
- Relace · input↓ 57%$0.35 → $0.15
- Relace · outputunchanged$3.50 → $3.50
- Relace · cache read↑ 50%$0.10 → $0.15
- Io Net · input↓ 29%$0.99 → $0.70
- Io Net · output↑ 11%$3.15 → $3.50
- Io Net · cache read↓ 17%$0.108 → $0.090
- GMICloud · input↓ 24%$1.74 → $1.32
- GMICloud · output↑ 14%$3.48 → $3.96
- GMICloud · cache read↓ 70%$0.145 → $0.044
- Wafer · input↓ 0.2%$0.395 → $0.394
- Wafer · output↓ 17%$4.20 → $3.49
- Wafer · cache read↓ 0.2%$0.316 → $0.315
| host | rate | was | now | change |
|---|---|---|---|---|
| Ionstream | input | $0.226 | $0.608 | ↑ 169% |
| Ionstream | output | $1.96 | $1.96 | — |
| Ionstream | cache read | $0.088 | $0.088 | — |
| Relace | input | $0.35 | $0.15 | ↓ 57% |
| Relace | output | $3.50 | $3.50 | — |
| Relace | cache read | $0.10 | $0.15 | ↑ 50% |
| Io Net | input | $0.99 | $0.70 | ↓ 29% |
| Io Net | output | $3.15 | $3.50 | ↑ 11% |
| Io Net | cache read | $0.108 | $0.090 | ↓ 17% |
| GMICloud | input | $1.74 | $1.32 | ↓ 24% |
| GMICloud | output | $3.48 | $3.96 | ↑ 14% |
| GMICloud | cache read | $0.145 | $0.044 | ↓ 70% |
| Wafer | input | $0.395 | $0.394 | ↓ 0.2% |
| Wafer | output | $4.20 | $3.49 | ↓ 17% |
| Wafer | cache read | $0.316 | $0.315 | ↓ 0.2% |
Ionstream input: ↑ 169%Relace input: ↓ 57%Relace cache read: ↑ 50%Io Net input: ↓ 29%Io Net output: ↑ 11%GMICloud input: ↓ 24%GMICloud output: ↑ 14%Wafer output: ↓ 17%
If you buy
Ionstream now costs more to run for input-heavy work, so check your usage mix and cache settings before your next top-up there.
The model
DeepSeek V4 Proopen the model →
Open weights1598.8B parameters1049K context28 hosts
intelligence36th of 168 writing32nd of 168 coding43rd of 168 agents31st of 55
DeepSeek V4 Pro is a large downloadable text model with a one-million-token request limit and strong mathematics scores. Its mixture-of-experts design keeps only 37 billion parameters active per token, making it more efficient to run than its total size suggests.
Source
Read on openrouter.ai · we published this on 29 September 2026.
read the original at openrouter.ai ↗