Day0 Support for Qwen3.8-Omni-Flash on LiteLLM

[misbahsy]

Support for Qwen3.8-Omni-Flash on LiteLLM

What's new:

Text, image, audio and video in, text out - reasons over audio and video together and calls tools, with a 1M-token context window and 131K max output.

Thinking is on by default at xhigh - reasoning tokens bill at the output rate. Set reasoning_effort to none, low or medium to cut it back. high and max both map to xhigh.

Pricing - $0.15 / MTok input, $0.016 cached input and $0.47 output (International). Qwen puts video input at about 89% cheaper than Qwen3.5-Omni-Plus.

Worth checking before you route traffic to it:

LiteLLM's DashScope provider defaults to the mainland China endpoint, so International accounts need api_base set as below. Audio goes in as a data:;base64, URL, not bare base64. And LiteLLM does not list reasoning_effort for DashScope yet, so send allowed_openai_params: ["reasoning_effort"] with it.

Get started, no redeploy needed:

Pricing lands in PR #41754. Reload your pricing in the UI under Models + Endpoints -> Price Data -> Reload Price Data (or POST /reload/model_cost_map as an admin), on any version v1.76.0 or newer.

model_list:
  - model_name: qwen3.8-omni-flash
    litellm_params:
      model: dashscope/qwen3.8-omni-flash
      api_base: https://dashscope-intl.aliyuncs.com/compatible-mode/v1
      api_key: os.environ/DASHSCOPE_API_KEY

Read the full guide → Day 0 Support: Qwen3.8-Omni-Flash

Misbah Syed LiteLLM AI Team