Qwen 3.8 27B (32K, thinking on)
Same model, more context, thinking switched on. Slower, and the extra room costs you speed.
Tested 2026-11-08
The same base model as my everyday local pick, configured the other way: 32K context with thinking left on. It is the same file on the same card, so the comparison is honest. Extra context means a bigger KV cache, which pushes the model closer to the edge of 16 GB and slows everything down. Thinking on means it spends tokens working the problem out before it answers.
I keep it installed for the jobs where the local model is the only option and the question is genuinely hard. In practice that is rare.
Pros
- 32K context holds a bigger chunk of a real project
- Thinking on gives noticeably better answers on multi-step problems
- Still free and still offline
Cons
- Slower per answer, and the wait is the whole cost here
- Closer to the edge of 16 GB of VRAM
- A bigger KV cache leaves less room for the model itself
What I tested
| Test | Result |
|---|---|
| Long document summary | TODO: compare against the 24K tag on the same document. |
| Coding task | Better than the 24K tag on a change that touched two files. Still slower than sending it to a cloud model. |
| Hard reasoning | TODO: record the question and result. |
| Image understanding | TODO: not tested yet. |
| Cost for a week of real use | $0.00 in tokens. The cost is time: answers take longer because the model thinks first. |
Bottom line: Keep it installed for hard questions you cannot send off the machine. For everything else the 24K tag is the better trade.
What this site is
I test AI tools myself — cloud models, local models, and the apps that manage them. Reviews only appear once I’ve run the same prompts through each tool on my own machine or account. No sponsorships, no affiliate links, no ads. More about how this works.
One email a week
What I tested this week, what it cost to run, and the one thing worth your time. No spam, unsubscribe in one click.