[BUG] OpenRouter DeepSeek V3.1 Terminus forces reasoning; cannot disable

[hannesrudolph]

Type

Bug

Problem / Value

When using DeepSeek V3.1 Terminus via OpenRouter, reasoning/thinking appears to be enabled by default and there’s no way to turn it off. This leads to worse answers for some tasks and increases token usage. Users expect to run the non‑reasoning variant but cannot opt out.

Context

• Affects users who select DeepSeek V3.1 Terminus under the OpenRouter provider.
• Observed in Settings and during chats after selecting the model.
• The rest of the product already supports configuring reasoning behavior; this specific model/provider path doesn’t let users disable it.

Reproduction

1. In Settings, choose OpenRouter as the provider and select DeepSeek V3.1 Terminus.
2. Start a chat or review the model controls.
3. There is no OFF option to disable reasoning/thinking; responses behave as if reasoning is on.

Expected result

• Clear ability to disable reasoning for this model on OpenRouter.
• When disabled, the model should not return or charge for reasoning tokens, and it should behave like the non‑reasoning variant.

Actual result

• Reasoning appears active by default for DeepSeek V3.1 Terminus on OpenRouter.
• No OFF control is available; users can’t opt out of reasoning behavior.

Constraints/Preferences

• Fix the behavior so reasoning can be turned off for this specific model/provider combination.
• Default to reasoning OFF for this model to avoid unexpected quality/cost impacts.

[roomote]

I'm exploring a potential fix for the DeepSeek V3.1 Terminus reasoning configuration issue and will open a PR to share progress.

[roomote]

Opened a PR with my attempt at addressing this issue: #8272

The fix adds explicit reasoning exclusion for DeepSeek V3.1 Terminus when reasoning is not enabled, following the same pattern used for Gemini models. This should allow users to disable reasoning (default OFF) while still being able to opt-in when needed.

Feedback welcome!

[aicodeking]

There’s also no gradation of reasoning effort here (low, medium, high, etc.). It’s simply either reasoning or non-reasoning. Therefore, the reasoning effort toggle should probably be removed as well.

[daniel-lxs]

OpenRouter simply doesn't provide enough information for us to determine exactly how reasoning is handled or which style it uses: https://openrouter.ai/docs/use-cases/reasoning-tokens. However, the reasoning options for OpenRouter should only be "always on" for models that have the property “isReasoningBudgetRequired." Either that or add another option to the reasoning effort dropdown to turn off reasoning.

Regardless, this fix should apply to all the OpenRouter models that have reasoning capabilities, but the reasoning itself can be turned off.