Anthropic Claude Sonnet 4 on Bedrock thinks it is Sonnet 3.5

It seems that Sonnet 4 thinks it is Sonnet 3.5. It also believes it's knowledge cutoff date is April 2024 instead of January 2025. This behaviour deviates from usage of Anthropic's Claude directly.

This can be reproduced in Bedrock playground or through the API. Minimal example using boto3:

import boto3
import json

# Use eu-south-2 region
bedrock = boto3.client(service_name="bedrock-runtime", region_name="eu-south-2")

body = json.dumps({
    "max_tokens": 256,
    "messages": [{"role": "user", "content": "Let me try this new Claude 4"}],
    "anthropic_version": "bedrock-2023-05-31"
})

# Use cross-region inference profile for Claude 4 Sonnet
response = bedrock.invoke_model(
    body=body, 
    modelId="eu.anthropic.claude-sonnet-4-20250514-v1:0" # same holds for inference profile of us
)

# Printing only the textual response
response_body = json.loads(response.get("body").read())
print(response_body['content'][0]['text'])

I'm actually Claude 3.5 Sonnet, not Claude 4! There isn't a Claude 4 released yet - you're using one of the current Claude 3.5 models from Anthropic. Is there something specific you'd like to try out or explore with Claude 3.5? I'm happy to help with questions, creative tasks, analysis, coding, or just have a conversation!

Claude.ai gives the following response for the same prompt:

Hello! Yes, you're chatting with Claude Sonnet 4 from the Claude 4 model family. I'm here to help with whatever you'd like to work on - whether that's answering questions, helping with analysis, creative projects, coding, or just having a conversation. What would you like to explore today?

Is Anthropic Claude Sonnet 4 on Bedrock suffering from an imposter syndrome, or a typo in it's system prompt or just unaware of any accurate information about itself?