You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Describe the bug
DeepSeek V4.1 Flash (deepseek-v4.1-flash) falls back to Zoo's default 200K context window when used with OpenCode Go, instead of its 1M context window.
To Reproduce
Select OpenCode Go as the provider.
Select DeepSeek V4.1 Flash.
Disable automatic context condensing.
Continue a conversation until the context reaches ~150K tokens.
Zoo starts omitting older messages.
Expected behavior
Zoo should use a 1M context window for DeepSeek V4.1 Flash.
Screenshots
Not applicable.
Video
Not applicable.
What version of zoo are you running
3.82.1
Additional context
The OpenCode Go API does not provide context-window metadata. Zoo's opencodeGoDefaultModelInfo uses a 200K fallback, while deepseek-v4-flash is explicitly configured with a 1M context window. deepseek-v4.1-flash appears to be missing from the model registry and therefore falls back to 200K.
As a suggestion for future models, it would be useful to allow users to override the context window value directly in the UI.
This would prevent similar issues when new models are released before their metadata is added to Zoo's model registry, without requiring a Zoo update for every new model.
Activity
carloslato commentedon Sep 15, 2026
As a suggestion for future models, it would be useful to allow users to override the context window value directly in the UI.
This would prevent similar issues when new models are released before their metadata is added to Zoo's model registry, without requiring a Zoo update for every new model.
PierrunoYT commentedon Sep 16, 2026
Working on this!