Qwen3.8 27B repriced across 3 hosts, from a 76% rise at DekaLLM (output) to an 11% cut at Ionstream (input)
Published 24 September 2026
Price move, per 1M tokens
- DekaLLM · input↓ 4%$0.100 → $0.096
- DekaLLM · output↑ 76%$2.50 → $4.40
- DekaLLM · cache read↑ 74%$0.050 → $0.087
- Ionstream · input↓ 11%$0.28 → $0.25
- Ionstream · outputunchanged$2.55 → $2.55
- Ionstream · cache readunchanged$0.10 → $0.10
- Wafer · input↓ 4%$0.099 → $0.095
- Wafer · outputunchanged$4.40 → $4.40
- Wafer · cache read↓ 4%$0.089 → $0.085
| host | rate | was | now | change |
|---|---|---|---|---|
| DekaLLM | input | $0.100 | $0.096 | ↓ 4% |
| DekaLLM | output | $2.50 | $4.40 | ↑ 76% |
| DekaLLM | cache read | $0.050 | $0.087 | ↑ 74% |
| Ionstream | input | $0.28 | $0.25 | ↓ 11% |
| Ionstream | output | $2.55 | $2.55 | — |
| Ionstream | cache read | $0.10 | $0.10 | — |
| Wafer | input | $0.099 | $0.095 | ↓ 4% |
| Wafer | output | $4.40 | $4.40 | — |
| Wafer | cache read | $0.089 | $0.085 | ↓ 4% |
DekaLLM input: ↓ 4%DekaLLM output: ↑ 76%Ionstream input: ↓ 11%Wafer input and cache read: ↓ 4%
If you buy
Output-heavy work at DekaLLM now costs noticeably more per call, so re-check your spend if most of your tokens are generated rather than read.
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 24 September 2026.
read the original at openrouter.ai ↗