Local model names can separate an application's chosen model from the original download name. An Ollama model copy adds a second local name while keeping the source available.

Ollama reuses the stored model blobs for both names rather than duplicating the weights. A copied name therefore does not provide an independent backup if the model store or its disk is lost.

The source must already be installed on the Ollama server that the CLI uses. Destination names share that server's model store, so an unused name avoids replacing a model an application already depends on.

Steps to copy an Ollama model:

  1. List the installed models before copying.
    $ ollama list
    NAME            ID              SIZE      MODIFIED
    qwen2.5:0.5b    6467b5fd1dd4    397 MB    5 hours ago

    The source here is qwen2.5:0.5b. If the intended source is absent, download it before copying. A connection error means the CLI cannot reach the server.
    Related: How to pull an Ollama model
    Related: How to start the Ollama server

    The copy can overwrite an existing destination name without a confirmation prompt. If example/qwen-copy:latest already appears in the list, use an unused destination name instead.

  2. Copy the source model to the new local name.
    $ ollama cp qwen2.5:0.5b example/qwen-copy
    copied 'qwen2.5:0.5b' to 'example/qwen-copy'

    The first argument is the installed source; the second is the destination. Omitting the destination tag selects latest. The example/ prefix is part of a local name and does not publish the model or require an account.

  3. Confirm that both names appear with the same model ID.
    $ ollama list
    NAME                        ID              SIZE      MODIFIED
    example/qwen-copy:latest    6467b5fd1dd4    397 MB    Less than a second ago
    qwen2.5:0.5b                6467b5fd1dd4    397 MB    5 hours ago

    The matching IDs identify the same model content under both names. The source remains installed; the displayed sizes do not mean that the weights occupy disk space twice.

  4. Inspect the copied model's metadata.
    $ ollama show example/qwen-copy
      Model
        architecture        qwen2
        parameters          494.03M
        context length      32768
        embedding length    896
        quantization        Q4_K_M
        template source     go_template
     
      Capabilities
        completion
        tools
    ##### snipped #####
  5. Run a prompt using the copied name.
    $ ollama run example/qwen-copy "What is 17 plus 25? Reply in one sentence."
    The sum of 17 and 25 is 42.

    The reply should give the sum 42, although wording can vary. A returned answer confirms that the copied name can load the model and process a prompt; metadata alone does not establish that. Both names remain available for later use.
    Related: How to run a model in Ollama

Share this guide