Qwen3.8 27B repriced across 4 hosts: Wafer output down 44%, Wafer input up 3%
Published 27 September 2026
Price move, per 1M tokens
- Wafer · inputunchanged$0.080 → $0.080
- Wafer · output↓ 44%$4.40 → $2.45
- Wafer · cache readunchanged$0.080 → $0.080
- DekaLLM · input↓ 16%$0.0960 → $0.0802
- DekaLLM · output↓ 43%$4.40 → $2.50
- DekaLLM · cache readunchanged$0.087 → $0.087
- Reka · input↓ 1%$0.081 → $0.080
- Reka · output↓ 43%$4.40 → $2.50
- Reka · cache read↓ 1%$0.081 → $0.080
- Darkbloom · input↓ 16%$0.0825 → $0.0690
- Darkbloom · outputunchanged$2.20 → $2.20
- Wafer · input↑ 3%$0.0803 → $0.0828
- Wafer · outputunchanged$2.45 → $2.45
- Wafer · cache readunchanged$0.080 → $0.080
- Reka · input↓ 1%$0.080 → $0.079
- Reka · outputunchanged$2.50 → $2.50
- Reka · cache read↓ 1%$0.080 → $0.079
| host | rate | was | now | change |
|---|---|---|---|---|
| Wafer | input | $0.080 | $0.080 | — |
| Wafer | output | $4.40 | $2.45 | ↓ 44% |
| Wafer | cache read | $0.080 | $0.080 | — |
| DekaLLM | input | $0.0960 | $0.0802 | ↓ 16% |
| DekaLLM | output | $4.40 | $2.50 | ↓ 43% |
| DekaLLM | cache read | $0.087 | $0.087 | — |
| Reka | input | $0.081 | $0.080 | ↓ 1% |
| Reka | output | $4.40 | $2.50 | ↓ 43% |
| Reka | cache read | $0.081 | $0.080 | ↓ 1% |
| Darkbloom | input | $0.0825 | $0.0690 | ↓ 16% |
| Darkbloom | output | $2.20 | $2.20 | — |
| Wafer | input | $0.0803 | $0.0828 | ↑ 3% |
| Wafer | output | $2.45 | $2.45 | — |
| Wafer | cache read | $0.080 | $0.080 | — |
| Reka | input | $0.080 | $0.079 | ↓ 1% |
| Reka | output | $2.50 | $2.50 | — |
| Reka | cache read | $0.080 | $0.079 | ↓ 1% |
Wafer output: ↓ 44%DekaLLM output: ↓ 43%Reka output: ↓ 43%Darkbloom input: ↓ 16%Wafer input: ↑ 3%Reka input and cache read: ↓ 1%
The model
Qwen3.8 27Bopen the model →
Open weights27.8B parameters262K context20 hosts
intelligence63rd of 168 writing93rd of 168 coding45th of 168 agents33rd of 55
Qwen3.8 27B is a downloadable model you can run yourself, and its measured strength is maths and reasoning. Agentic recovery and tool use sit near the bottom of the field, so it suits a solver rather than an agent that has to get itself back on track.
Source
Read on openrouter.ai · we published this on 27 September 2026.
read the original at openrouter.ai ↗