"thinking.type.enabled" is not supported for this model. Use "thinking.type.adaptive" and "output_config.effort" to control thinking behavior.

Raw SDK

Remove budget_tokens. The migration guide's after request uses thinking type adaptive and output_config.effort. The guide's example uses high, and it also lists max, xhigh, medium, and low.

client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=16000,
    thinking={"type": "enabled", "budget_tokens": 10000},
    messages=[{"role": "user", "content": "Hello"}],
)

client.messages.create(
    model="claude-sonnet-5-5",
    max_tokens=16000,
    thinking={"type": "adaptive"},
    output_config={"effort": "high"},
    messages=[{"role": "user", "content": "Hello"}],
)

Source: Claude Sonnet 5.5 migration guide, checked October 20, 2026.

LiteLLM

Do not send thinking type enabled with budget_tokens. LiteLLM documents a thinking parameter on Anthropic calls. Pass the adaptive object from the migration guide. LiteLLM's reasoning_effort map is documented for Claude 4.6 and 4.7, not for claude-sonnet-5-5.

import litellm
response = litellm.completion(
    model="anthropic/claude-sonnet-5-5",
    messages=[{"role": "user", "content": "Hello"}],
    thinking={"type": "adaptive"},
)

Source: LiteLLM Anthropic provider, checked October 20, 2026.

Vercel AI SDK

The AI SDK Anthropic docs show thinking type adaptive plus effort on providerOptions.anthropic. They also say the provider converts thinking type enabled with budgetTokens to adaptive for claude-opus-5-5, claude-fable-5, and claude-fable-5-1. They do not name claude-sonnet-5-5 for that conversion, so set adaptive yourself.

import { anthropic, type AnthropicProviderOptions } from "@ai-sdk/anthropic";
import { generateText } from "ai";

const { text } = await generateText({
  model: anthropic("claude-sonnet-5-5"),
  prompt: "Hello",
  providerOptions: {
    anthropic: {
      thinking: { type: "adaptive" },
      effort: "high",
    } satisfies AnthropicProviderOptions,
  },
});

Source: AI SDK Anthropic provider, checked October 20, 2026.

Questions