🧠 DeepSeek-V4.1-Flash — Chat Demo

Multimodal MoE · 552B backbone params · 1M-token context · KV cache 890 bytes/token · Model card

This demo runs fully in your browser. It calls Hugging Face Inference Providers directly — no server, no GPU, nothing is stored except your token in your own browser.

No token saved. Create a fine-grained token (only Inference Providers permission) at huggingface.co/settings/tokens.
Your token is stored only in this browser’s localStorage and sent only to Hugging Face. Use a token with only Inference Providers permission — never a full-write token.
If a provider rejects a request (e.g. image input or reasoning_effort), try another provider from the dropdown.