The official Ollama library now carries deepseek-v4-flash, which makes local and cloud-proxied experiments much less painful than hand-rolling GGUF downloads.

The detail that matters for agent users: OpenClaw can be launched directly against the model with `ollama launch openclaw --model deepseek-v4-flash:cloud`. That one line replaces a manual gateway configuration and is currently the fastest way to test Flash as an OpenClaw default route.

Local runs still have real hardware requirements — quantized builds start around 33 GB of VRAM. Our local setup guide breaks down the three hardware tiers and which quantization to pick for each.

Sources and publication record

Source material is linked for readers who want to verify the underlying announcement or documentation.