Similar model names can hide differences in size, precision, and supported features. Ollama's model details describe the installed variant before you choose it for an application.
A model's stored context length describes its capacity rather than the context allocated to a running request. Its embedding length is an architectural property and does not establish support for the embeddings API.
An Ollama server must be running with the model already downloaded. The full model name, including its tag, identifies which installed variant to inspect; metadata queries leave its files unchanged.
Steps to show Ollama model details:
- Show the summary for the installed qwen2.5:0.5b model.
$ ollama show qwen2.5:0.5b Model architecture qwen2 parameters 494.03M context length 32768 embedding length 896 quantization Q4_K_M template source go_template Capabilities completion tools ##### snipped #####The summary identifies the architecture, parameter count and quantization of this variant. Capabilities lists the features Ollama reports for it, such as completion and tools. Another installed model may report different values. If the name is unknown, Ollama returns a model-not-found error instead of metadata.
Related: How to list Ollama models - Display the model's generated Modelfile.
$ ollama show --modelfile qwen2.5:0.5b
The FROM line points to the local model blob, while TEMPLATE describes how messages are formatted. SYSTEM and LICENSE contain the system message and license when supplied by the model. PARAMETER entries appear when model-specific settings are stored; their absence does not mean Ollama has no runtime defaults. The generated blob path belongs to that server's model store and can differ between installations.
- Confirm the same model's metadata through the Ollama API.
$ curl -fsS http://localhost:11434/api/show -d '{"model":"qwen2.5:0.5b"}'
The JSON response includes details.family, details.parameter_size and details.quantization_level, which are qwen2, 494.03M and Q4_K_M for this model. Its capabilities array contains completion and tools, and template contains the prompt template seen in the Modelfile. These matching fields confirm that the API and CLI describe the same installed variant. If OLLAMA_HOST directs the CLI to another server, the API URL must address that server too. The -f option makes an HTTP error return a failing curl exit status.