Amazon web services – SQS lambda trigger not working as expected

The scenario is this: When my process start I send around 3k messages to a queue in SQS. From there I have a lambda that I want to pick 500 of those messages and process them, then take the next 500 messages and so on until the queue is empty. Since the lambda that processes the messages has to talk to a very slow API I can’t use concurrent lambda executions, otherwise I will reach the limits on that API.
The problem that I’m seeing is that sometimes the lambda starts picking 500 messages but in the next execution picks less than 200 and the next execution it picks even less and so on, instead of picking 500 everytime.
Maybe I misunderstanding something in the settings. How can I achieve this behaviour?
These are my settings:

Lambda:
ProcessMessages:
    Type: AWS::Serverless::Function
    Properties:
      Timeout: 600
      CodeUri: process_messages/
      Role: !Sub ${ConsumerLambdaRole.Arn}
      Handler: app.lambda_handler
      Runtime: python3.8
      ReservedConcurrentExecutions: 1
      Architectures:
        - x86_64
Queue:
  ProcessMessagesQueue:
    Type: AWS::SQS::Queue
    Properties:
      QueueName: 'processMessages'
      DelaySeconds: 0
      VisibilityTimeout: 900
      RedrivePolicy:
        deadLetterTargetArn: !GetAtt DLmessages.Arn
        maxReceiveCount: 10

EventSource:
  SendMessageToLambda:
    Type: AWS::Lambda::EventSourceMapping
    Properties:
      BatchSize: 500
      Enabled: true
      EventSourceArn: !GetAtt ProcessMessagesQueue.Arn
      FunctionName: !GetAtt ProcessMessages.Arn
      FunctionResponseTypes: 
        - "ReportBatchItemFailures"
      MaximumBatchingWindowInSeconds: 20