feat(opencode): Add none effort level to deepseek-v4 models

[martinmr]

Issue for this PR

Closes #24610

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Adds a none variant to deepseek v4 models that enables non-thinking mode. These models tend to overthink a lot, so this feature is a quality of life improvement.

How did you verify your code works?

Ran it locally, verified that none option is there, and that it does not produce any thinking traces.

Screenshots / recordings

N/A

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

[martinmr]

No idea what the other changes are here. My changes were only to transform.ts. I am not sure how to resolve the conflicts so I'll let a maintainer do it.

[LuisPeregrina]

Can any maintainer add dumb commit? Seems github had a hiccup and didn't continue with CI

[martinmr]

Can any maintainer add dumb commit? Seems github had a hiccup and didn't continue with CI

It’s not a hiccup. Maintainers have to approve CI runs.

[tanishqvec]

Your payload is the right one for api.deepseek.com, but I don't think variants() is reached for these models: provider.ts:1310 does ProviderTransform.reasoningVariants(model, base) ?? ProviderTransform.variants(base), and because models.dev declares reasoning_options for deepseek-v4-*, reasoningVariants returns low/high/max and short-circuits the fallback. Worth checking against your build in case I've misread it.

Your model.providerID === "deepseek" guard was the right instinct — I counted ~63 models across ~30 providers that would match on the model id alone, and zen turns out to need a different payload again (it ignores thinking and accepts reasoning_effort: "none").

I opened #46181, which moves this into reasoningVariants and keeps your provider scoping. Happy to close mine if you'd prefer to move yours instead.