Skip to content

claude sonnet "thinking"? #1106

Closed
Closed

Description

@jubishop

im not sure how to make open code use the claude sonnet thinking model with my claude pro plan? im using sonnet fine, but unclear how to make it "thinking".?

Activity

  1. almirsarajcic commented on Jul 18, 2025

    @almirsarajcic
    Contributor

    Have you tried using the word "think" in your prompt?

    Ask Claude to make a plan for how to approach a specific problem. We recommend using the word "think" to trigger extended thinking mode, which gives Claude additional computation time to evaluate alternatives more thoroughly. These specific phrases are mapped directly to increasing levels of thinking budget in the system: "think" < "think hard" < "think harder" < "ultrathink." Each level allocates progressively more thinking budget for Claude to use.

    https://www.anthropic.com/engineering/claude-code-best-practices

    Here's an example: https://opencode.ai/s/BT8v9iuG (in the middle, I ask it to think hard).

    But yeah, it's not clear when it's using the thinking mode.
    In Claude Code there's an indicator * Thinking. Maybe opencode should show thinking... instead of working... in that case.

  2. jubishop commented on Jul 18, 2025

    @jubishop
    Author

    thanks for the info!

  3. self-assigned this
    on Jul 18, 2025
  4. thdxr commented on Jul 18, 2025

    @thdxr
    Member

    we don't support this yet but we're talking about how to implement it

  5. nerdo commented on Jul 30, 2025

    @nerdo

    @thdxr To clarify, do you mean that extended thinking modes do not yet work, or is it opencode exposing Claude's thought process a-la * Thinking in Claude Code that isn't working... or both?

    I'd love the extended thinking feedback, but I kind of rely on extended thinking for a lot of my work so I need to understand whether or not it's actually working or if I should wait until there is some basic support for extended thinking modes.

  6. IV2KBMoFxYIA commented on Aug 21, 2025

    @IV2KBMoFxYIA

    @thdxr To clarify, do you mean that extended thinking modes do not yet work, or is it opencode exposing Claude's thought process a-la * Thinking in Claude Code that isn't working... or both?

    I'd love the extended thinking feedback, but I kind of rely on extended thinking for a lot of my work so I need to understand whether or not it's actually working or if I should wait until there is some basic support for extended thinking modes.

    @thdxr any reply on this? it's a pretty important distinction.

  7. rekram1-node commented on Aug 21, 2025

    @rekram1-node
    Contributor

    We now have thinking blocks @IV2KBMoFxYIA, we need better documentation but w/ anthropic models we do not have thinking enabled by default, I will comment a config here of how to do it and then I will probably update docs too

    /thinking will make thinking blocks show

    they look like:

    Image
  8. nerdo commented on Aug 22, 2025

    @nerdo

    Please do so! I've been waiting for this! Better yet, just update the docs and link it here.

  9. rekram1-node commented on Aug 25, 2025

    @rekram1-node
    Contributor

    @nerdo here is how you can set it in your opencode.json for anthropic:

    "provider": {
        "anthropic": {
          "name": "Anthropic",
          "models": {
            "claude-sonnet-4-20250514": {
              "options": {
                "thinking": {
                  "type": "enabled",
                  "budgetTokens": 16000
                }
              }
            }
          }
        }
      }
    
  10. nerdo commented on Aug 26, 2025

    @nerdo
  11. rekram1-node commented on Aug 26, 2025

    @rekram1-node
    Contributor

    I recall Dax mentioning wanting to add a better DX for doing it dynamically but rn it is by config only

  12. iamhenry commented on Nov 5, 2025

    @iamhenry

    @rekram1-node is the thinking block missing from the new v1.0+?

  13. rekram1-node commented on Nov 5, 2025

    @rekram1-node
    Contributor

    @iamhenry there is one:

    Image

    One may argue it is non obvious if it is an assistant message vs reasoning one.

    Here is share link too: https://opencode.ai/s/z6QQSSh9

  14. iamhenry commented on Nov 5, 2025

    @iamhenry

    Here is share link too: https://opencode.ai/s/z6QQSSh9

    Ah, yes. It just wasn't clear I guess. Thanks!

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

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions