You tried to run a model and got:

Error: pull model manifest: file does not exist

Or:

Error: model "my-model" not found, try pulling it first

Here’s every reason this happens and how to fix each one.

Fix 1: Check the model name

The most common cause — a typo or wrong model name:

# Wrong (common mistakes)
ollama run llama-4          # Hyphen instead of no separator
ollama run deepseek-r1-14b  # Wrong format
ollama run qwen3.5-27b      # Wrong format

# Right
ollama run llama4-scout
ollama run deepseek-r1:14b
ollama run qwen3.5:27b

The format is model-name:tag. The colon separates the model from the size/quantization tag. Search the Ollama library for the exact name.

# Search for available models
ollama list  # Shows locally downloaded models
# Check ollama.com/library for the full catalog

Fix 2: Pull before running

Ollama doesn’t auto-download models on run if they’re not in the library:

# Pull first, then run
ollama pull qwen3.5:27b
ollama run qwen3.5:27b

# Or pull + run in one step (works for library models)
ollama run qwen3.5:27b  # Auto-pulls if in the official library

If the model isn’t in Ollama’s official library (e.g., a community model or custom GGUF), you need to import it manually.

Fix 3: Import a GGUF model

For models not in the Ollama library (like Jais or custom fine-tunes):

# Download the GGUF file
huggingface-cli download TheBloke/model-name-GGUF model.Q5_K_M.gguf --local-dir ./

# Create a Modelfile
cat > Modelfile << 'EOF'
FROM ./model.Q5_K_M.gguf
PARAMETER temperature 0.7
PARAMETER num_ctx 4096
EOF

# Import into Ollama
ollama create my-custom-model -f Modelfile

# Now it works
ollama run my-custom-model

Fix 4: Network/registry issues

If ollama pull fails:

# Check if Ollama's registry is reachable
curl -s https://registry.ollama.ai/v2/ | head -5

# If behind a proxy
export HTTPS_PROXY=http://your-proxy:8080
ollama pull qwen3.5:27b

# If DNS issues
# Try Google DNS
echo "nameserver 8.8.8.8" | sudo tee /etc/resolv.conf

Fix 5: Corrupted download

If a pull was interrupted, the model might be partially downloaded:

# Remove the corrupted model
ollama rm model-name

# Re-pull
ollama pull model-name

Fix 6: Docker-specific issues

If running Ollama in Docker, the model storage might not be mounted:

# docker-compose.yml — mount the model directory
services:
  ollama:
    image: ollama/ollama
    volumes:
      - ollama_data:/root/.ollama  # Persist models between restarts
volumes:
  ollama_data:

Without the volume mount, models are lost every time the container restarts.

Quick reference: popular model names

Model	Ollama command
Qwen 3.5 27B	ollama run qwen3.5:27b
DeepSeek R1 14B	ollama run deepseek-r1:14b
Llama 4 Scout	ollama run llama4-scout
Gemma 4 9B	ollama run gemma4:9b
Phi-4 14B	ollama run phi4:14b
Mistral Large 2	ollama run mistral-large
CodeStral	ollama run codestral

For the full list of models and which ones are best for coding, see our best Ollama models guide.

Related: Ollama Complete Guide · Ollama Troubleshooting Guide · Ollama Out of Memory Fix · Best Ollama Models for Coding · How to Run DeepSeek Locally · How to Run Qwen 3.5 Locally