Skip to content
Docs

AI Gateway Amazon Bedrock Reasoning

Amazon Bedrock exposes Anthropic Claude reasoning through model-creator-specific provider options. Configuration depends on the model:

  • Adaptive reasoning: Set reasoningConfig: { type: 'adaptive', maxReasoningEffort }. Available on Claude 4.6 and later. Required on Claude Opus 4.7 and later, where the legacy type: 'enabled' mode returns a 400 error.
  • Manual reasoning: Set reasoningConfig: { type: 'enabled', budgetTokens: N } for a fixed token budget. Available on Claude 4.6 and earlier (deprecated on 4.6, removed on Claude Opus 4.7 and later).

Supported models

See the AI SDK Amazon Bedrock provider reference for supported reasoning options and response metadata.

Use GET /v1/models to discover a model's reasoning controls. Then use model endpoints or the Bedrock model filter to check which models Bedrock serves. The model catalog's creator slug remains unchanged when you route through Bedrock.

Keep the adaptive and legacy request shapes above separate. Catalog metadata describes model controls; the serving provider can impose additional restrictions.

Getting started

Top-level reasoning option

The AI SDK 7 top-level reasoning option works with Bedrock-hosted Claude models without provider-specific configuration. On Claude 4.6 and later it maps to adaptive reasoning at the corresponding effort level:

top-level-reasoning.ts
import { generateText } from 'ai';
 
const result = await generateText({
  model: 'anthropic/claude-opus-5',
  prompt: 'How many "r"s are in the word "strawberry"?',
  reasoning: 'high',
});
 
console.log(result.text);

This is especially useful with provider fallbacks: the same setting applies whether Anthropic, Bedrock, or Vertex serves the request. Use providerOptions.bedrock when you need an exact token budget on older models. If you set reasoningConfig in providerOptions, it takes precedence over the top-level reasoning value.

Adaptive reasoning (Claude 4.6 and later)

bedrock-adaptive.ts
import { generateText } from 'ai';
 
const result = await generateText({
  model: 'anthropic/claude-opus-5',
  prompt: 'How many "r"s are in the word "strawberry"?',
  providerOptions: {
    bedrock: {
      reasoningConfig: { type: 'adaptive', maxReasoningEffort: 'max' },
    },
  },
});
 
console.log(result.reasoningText);
console.log(result.text);

Manual reasoning (Claude 4.6 and earlier)

For pre-4.7 models, use type: 'enabled' with a budgetTokens value. This is the only reasoning mode supported on Claude 4.5 and earlier; on Claude 4.6 it works but is deprecated; on Claude Opus 4.7 and later it returns a 400 error.

bedrock-manual.ts
import { generateText } from 'ai';
 
const result = await generateText({
  model: 'anthropic/claude-sonnet-4.5',
  prompt: 'How many people will live in the world in 2040?',
  providerOptions: {
    bedrock: {
      reasoningConfig: { type: 'enabled', budgetTokens: 2048 },
    },
  },
});
 
console.log(result.reasoningText);
console.log(result.text);

Other API formats

You can configure reasoning without the AI SDK through the gateway's OpenAI-compatible endpoints. Use providerOptions.gateway.order to route the request to Bedrock; AI Gateway maps the reasoning effort level to Bedrock's reasoning configuration:

reasoning-chat-completions.ts
import OpenAI from 'openai';
 
const openai = new OpenAI({
  apiKey: process.env.AI_GATEWAY_API_KEY,
  baseURL: 'https://ai-gateway.vercel.sh/v1',
});
 
const completion = await openai.chat.completions.create({
  model: 'anthropic/claude-opus-5',
  messages: [
    {
      role: 'user',
      content: 'How many "r"s are in the word "strawberry"?',
    },
  ],
  // AI Gateway extension fields are not included in the upstream SDK types.
  ...{
    reasoning: {
      effort: 'high',
    },
  },
  ...{
    providerOptions: {
      gateway: {
        order: ['bedrock'],
      },
    },
  },
});
 
const message = completion.choices[0].message;
console.log(
  'Reasoning:',
  'reasoning' in message ? message.reasoning : undefined,
);
console.log('Answer:', completion.choices[0].message.content);
reasoning_chat_completions.py
import os
from openai import OpenAI
 
client = OpenAI(
    api_key=os.getenv('AI_GATEWAY_API_KEY'),
    base_url='https://ai-gateway.vercel.sh/v1'
)
 
completion = client.chat.completions.create(
    model='anthropic/claude-opus-5',
    messages=[
        {
            'role': 'user',
            'content': 'How many "r"s are in the word "strawberry"?'
        }
    ],
    extra_body={
        'reasoning': {
            'effort': 'high'
        },
        'providerOptions': {
            'gateway': {
                'order': ['bedrock']
            }
        }
    }
)
 
print('Reasoning:', completion.choices[0].message.reasoning)
print('Answer:', completion.choices[0].message.content)
reasoning-chat-completions.sh
curl https://ai-gateway.vercel.sh/v1/chat/completions \
  -H "Authorization: Bearer $AI_GATEWAY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-opus-5",
    "messages": [
      {
        "role": "user",
        "content": "How many \"r\"s are in the word \"strawberry\"?"
      }
    ],
    "reasoning": {
      "effort": "high"
    },
    "providerOptions": {
      "gateway": {
        "order": ["bedrock"]
      }
    }
  }'

Parameters

Adaptive reasoning (Claude 4.6 and later)

ParameterTypeDescription
typestringSet to 'adaptive'
maxReasoningEffortstringA level supported by the model's catalog entry and the Bedrock provider options

Manual reasoning (Claude 4.6 and earlier)

ParameterTypeDescription
typestringSet to 'enabled'
budgetTokensnumberToken budget for reasoning. Minimum: 1,024. Maximum: 64,000
Last updated September 8, 2026
Previous
Google / Vertex
Next
Web Search

Was this helpful?

supported.