sgl-project/sglang · error · ValueError

thinking.budget_tokens is not allowed when thinking.type is

Error message

thinking.budget_tokens is not allowed when thinking.type is 'adaptive'

What it means

Raised by the Anthropic-compatible endpoint's validator when thinking.type='adaptive' (SGLang's adaptive thinking mode) and a budget_tokens value is also supplied. Adaptive thinking decides the budget itself, so an explicit budget_tokens is contradictory and rejected with a 400.

Source

Thrown at python/sglang/srt/entrypoints/anthropic/protocol.py:308

            if self.budget_tokens < 1024:
                raise ValueError(
                    "thinking.budget_tokens must be >= 1024 "
                    "(got {})".format(self.budget_tokens)
                )
        elif self.type == "disabled":
            if self.budget_tokens is not None:
                raise ValueError(
                    "thinking.budget_tokens is not allowed when "
                    "thinking.type is 'disabled'"
                )
            if self.display is not None:
                raise ValueError(
                    "thinking.display is not allowed when "
                    "thinking.type is 'disabled'"
                )
        elif self.type == "adaptive":
            if self.budget_tokens is not None:
                raise ValueError(
                    "thinking.budget_tokens is not allowed when "
                    "thinking.type is 'adaptive'"
                )
        return self


class AnthropicTaskBudget(BaseModel):
    """Claude 4.7 ``output_config.task_budget`` — soft hint, not a hard cap.

    Mirrors ``BetaTokenTaskBudgetParam`` in the Anthropic SDK: ``total``
    and ``type`` are required; ``remaining`` is the client-tracked
    countdown used for compaction. The hard cap on generation is still
    ``max_tokens``; we never enforce ``task_budget`` ourselves.
    """

    type: Literal["tokens"]
    total: int = Field(gt=0)
    remaining: Optional[int] = Field(default=None, ge=0)

View on GitHub (pinned to 0132848349)

Solutions

1. Remove budget_tokens when using thinking.type='adaptive'
2. Use thinking.type='enabled' with budget_tokens if you need an explicit cap

Example fix

// before
{"thinking": {"type": "adaptive", "budget_tokens": 2048}}
// after
{"thinking": {"type": "adaptive"}}

if thinking["type"] == "adaptive":
    thinking = {"type": "adaptive"}

When it happens

Trigger: POST /v1/messages with {"thinking": {"type": "adaptive", "budget_tokens": 2048}}.

Common situations: Switching a request template from 'enabled' to 'adaptive' without cleaning out budget_tokens; assuming adaptive accepts an upper bound like some providers' adaptive modes do.

Related errors

• thinking.budget_tokens must be >= 1024 (got {})
• thinking.budget_tokens is not allowed when thinking.type is
• thinking.display is not allowed when thinking.type is 'disab
• Model is required
• max_tokens must be positive

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/a5f9150eafd6d722. Report an issue: GitHub.