Skip to content

llm-council: no way to pass reasoning_effort to OpenAI-compatible or thinking budget to Anthropic #96

Description

@23r2efewvcs

Enhancement

Modern reasoning models (OpenAI o1/o3, DeepSeek R1/V4, Claude with thinking) accept a reasoning_effort parameter that controls how much compute they spend on internal reasoning. Without it, models either:

  • Use a default that's too high (wasting tokens/money on simple queries)
  • Use a default that's too low (truncating reasoning on complex queries)

There's no way to pass this from the CLI.

Current behavior

callOpenAICompat sends { model, messages, max_tokens, temperature: 1 } — no reasoning parameter.
callAnthropic sends { model, max_tokens, system, messages } — no thinking budget.

Proposed fix

  • Add --reasoning-effort low|medium|high CLI flag
  • In callOpenAICompat: add reasoning: { effort } to payload when set
  • In callAnthropic: add thinking: { type: 'enabled', budget_tokens } to payload when set (budget derived from effort level)

Usage after fix

node council.js run "<query>" --models "..." --chairman "..." --reasoning-effort high

Activity

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

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions