Repository navigation
[FEATURE]:increase session limit of deepseek V4.1 flash the free model #56034
Description
Feature hasn't been suggested before.
- I have verified this feature I'm about to request hasn't been suggested before.
Describe the enhancement you want to request
the 200k tokens is to short , somtimes make compactions 5 times in one task
Activity
opencode-agent commentedon Oct 16, 2026
Thanks for the request, @ghassenTn! You're asking for a larger context window than 200k tokens for the free DeepSeek V4.1 Flash model, because long tasks currently hit several compactions.
The context limit for this model is set by the OpenCode Zen service, not by the app, so a plugin or config change can't raise it. The maintainers will consider it, but I can't promise anything.
Related issues you may want to follow:
recent, which can make the context bigger than before compaction #51818: compaction can sometimes leave the context bigger than before, which could explain compaction happening more often than expectedIf you need a longer context now, you can pick a model with a larger context window from the model picker.
Sshmoob commentedon Oct 18, 2026
Long context is so much cheaper with this model (because of CSA 2) that I might bet that these constant compactions that are therefore constantly breaking the cache are costing them more than if they just increased the context window. I'm not the one paying for it though, so lets be patient and wait to see what the devs have to say.
neriousy commentedon Oct 21, 2026
Thanks for the request. Which models we offer in Zen and Go, and their limits, depends on pricing, capacity and our own testing, so we don't track individual requests like this on GitHub. The free DeepSeek V4.1 Flash is limited to 200K context. If you need more, the paid DeepSeek V4.1 Flash in Zen (https://opencode.ai/v2/docs/console/models) and Go (https://opencode.ai/v2/docs/console/go) supports the model's full 1M context. Closing as not planned.