Qwen3.8 27B repriced across 5 hosts: DekaLLM input down 39%, Ionstream input up 37%
Published 30 September 2026
Price move, per 1M tokens
- DekaLLM · input↓ 39%$0.080 → $0.049
- DekaLLM · output↑ 20%$2.50 → $3.00
- DekaLLM · cache read↓ 50%$0.040 → $0.020
- Ionstream · input↑ 37%$0.145 → $0.198
- Ionstream · outputunchanged$2.55 → $2.55
- Ionstream · cache readunchanged$0.085 → $0.085
- AkashML · input↓ 10%$0.225 → $0.203
- AkashML · output↓ 10%$1.98 → $1.78
- AkashML · cache readunchanged$0.045 → $0.045
- Wafer · inputunchanged$0.025 → $0.025
- Wafer · output↓ 1%$4.40 → $4.35
- Wafer · cache readunchanged$0.020 → $0.020
- Reka · inputunchanged$0.025 → $0.025
- Reka · output↓ 1%$4.40 → $4.35
- Reka · cache read↓ 22%$0.0199 → $0.0155
| host | rate | was | now | change |
|---|---|---|---|---|
| DekaLLM | input | $0.080 | $0.049 | ↓ 39% |
| DekaLLM | output | $2.50 | $3.00 | ↑ 20% |
| DekaLLM | cache read | $0.040 | $0.020 | ↓ 50% |
| Ionstream | input | $0.145 | $0.198 | ↑ 37% |
| Ionstream | output | $2.55 | $2.55 | — |
| Ionstream | cache read | $0.085 | $0.085 | — |
| AkashML | input | $0.225 | $0.203 | ↓ 10% |
| AkashML | output | $1.98 | $1.78 | ↓ 10% |
| AkashML | cache read | $0.045 | $0.045 | — |
| Wafer | input | $0.025 | $0.025 | — |
| Wafer | output | $4.40 | $4.35 | ↓ 1% |
| Wafer | cache read | $0.020 | $0.020 | — |
| Reka | input | $0.025 | $0.025 | — |
| Reka | output | $4.40 | $4.35 | ↓ 1% |
| Reka | cache read | $0.0199 | $0.0155 | ↓ 22% |
DekaLLM input: ↓ 39%DekaLLM output: ↑ 20%Ionstream input: ↑ 37%AkashML input and output: ↓ 10%Wafer output: ↓ 1%Reka output: ↓ 1%
If you buy
Ionstream now costs more to prompt, so if your workload is input-heavy, re-check your budget before your next top-up; output-heavy use is less affected.
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 30 September 2026.
read the original at openrouter.ai ↗