Skip to content

[FEATURE]: Add V2 reasoning option support #34488

Closed

Description

@opencode-agent

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

Add/repair V2 reasoning/thinking option support across model metadata, TUI selection, and runtime requests.

Context from the V2 beta discussion:

  • Reasoning options need a V2 pass.
  • The UI should expose valid thinking/reasoning levels for models that support them.
  • Runtime requests should send the correct provider-specific fields.

Expected scope:

  • Define how V2 model metadata represents reasoning/thinking options.
  • Surface valid options to the TUI/client.
  • Persist/session-apply selected reasoning options where intended.
  • Map selected options into provider request payloads.
  • Add tests for at least one OpenAI-style and one Anthropic-style reasoning model path if applicable.

Activity

  1. github-actions commented on Jun 29, 2026

    @github-actions
    Contributor

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

  2. opencode-agent commented on Jul 2, 2026

    @opencode-agent
    ContributorAuthor

    Tracker update from V2 TODO LIST audit (2026-07-02):

    Active implementation path:

    Current user symptom still reported: Anthropic thinking levels are not selectable in V2. Treat this issue as not complete until Anthropic thinking variants appear in the TUI and selected variants lower to the provider request correctly.

  3. opencode-agent commented on Jul 2, 2026

    @opencode-agent
    ContributorAuthor

    Tracker update from V2 TODO LIST (2026-07-02):

    @rekram1-node reports reasoning variants should now be properly fixed in V2 after merging the relevant PR. Please re-test:

    • Anthropic thinking-level selection in TUI (/variants / picker)
    • selected variant identity persistence
    • request lowering for at least Anthropic and OpenAI-style reasoning options

    If those pass, this issue can be closed or narrowed to any remaining model/provider-specific gaps.

  4. opencode-agent commented on Jul 2, 2026

    @opencode-agent
    ContributorAuthor

    Tracker update from V2 TODO LIST (2026-07-02): @rekram1-node reports variants/reasoning should now work properly in V2 after merging the relevant PR(s), including #34726. Next action: verify in the V2 TUI that Anthropic thinking-level selection is visible and that selected variants lower correctly into provider requests. If verified, close this issue or narrow it to any remaining provider-specific cases.

  5. opencode-agent commented on Jul 5, 2026

    @opencode-agent
    ContributorAuthor

    Consolidated #34487 here. This issue now also owns the generic variant path: client/TUI selection must persist the correct variant identity and apply it to the provider request, not only expose reasoning metadata.

  6. opencode-agent commented on Jul 8, 2026

    @opencode-agent
    ContributorAuthor

    Completed by #34618 and #34726. #34618 added typed parsing for models.dev reasoning options; #34726 derived provider-backed variants, exposed them through the V2 model/client surfaces, and applied the selected provider options in V2 runtime requests, with Anthropic and OpenAI protocol coverage. Closing as completed.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions