DeepSeek V4 output limit is capped at 32K #38236

Activity

1. github-actions

   assigned

   rekram1-node

   on Jul 22, 2026
2. github-actions commented on Jul 22, 2026

   

   github-actionsBot

   on Jul 22, 2026 – with GitHub Actions

   Contributor

   More actions

   This issue might be a duplicate of existing issues. Please check:

   • Bug: limit.output in config is silently capped at 32k; OPENCODE_EXPERIMENTAL_OUTPUT_TOKEN_MAX is a poor workaround #29363: limit.output is silently capped at 32k with the same root cause — ProviderTransform.maxOutputTokens() applying the 32K global default regardless of model-declared limits, and OPENCODE_EXPERIMENTAL_OUTPUT_TOKEN_MAX being the only workaround
   • [Bug] OpenCode Go deepseek-v4-pro sends an invalid max_completion_tokens value #37947: DeepSeek V4 Pro sending an invalid max_completion_tokens value — related DeepSeek token limit handling issue

   If your issue is distinct from the above (e.g. you specifically want model-declared output limits to be respected automatically without any config override), please clarify the difference in a comment so maintainers can track it separately.
3. punkisnotdead3 commented on Sep 9, 2026

   

   punkisnotdead3

   on Sep 9, 2026

   Author

   More actions

   Clarifying the distinction from #29363: this issue targets the no-override path for DeepSeek V4. The model metadata already declares a 384K/393,216 output limit, but the shipped dev path still replaces it with the generic 32K default in ProviderTransform.maxOutputTokens(). #38232 deliberately keeps 32K for other model families and preserves OPENCODE_EXPERIMENTAL_OUTPUT_TOKEN_MAX, so the change is narrow and low-risk while the broader v2 rewrite is pending.

   I synced #38232 with current dev in 31f8b4195a, resolved the conflict, and revalidated the focused tests and typecheck.

   👍React with 👍1Reacted by FyiurAmron
4. neriousy commented on Oct 1, 2026

   

   neriousy

   on Oct 1, 2026

   Member

   More actions

   Closing as a duplicate of #29363. DeepSeek is one case of the general per-step output cap tracked in #29363.
5. neriousy

   closed this as a duplicateof Bug: limit.output in config is silently capped at 32k; OPENCODE_EXPERIMENTAL_OUTPUT_TOKEN_MAX is a poor workaround #29363 on Oct 1, 2026