Qwen3.5-Omni

What is Qwen3.5-Omni?

Qwen3.5-Omni is one model that listens, watches and speaks: it reads text, images, audio and video together and replies in text or a spoken voice, instead of chaining separate speech-to-text, language and text-to-speech models. It comes as Plus and Flash on Alibaba Cloud Model Studio, each with a Realtime version for live voice. Qwen has not released the weights; its technical report says the model runs to hundreds of billions of parameters.

It is still sold, but it is no longer Alibaba's newest omni model. Model Studio now sends text-only analysis to Qwen3.8-Omni-Flash and live voice to Qwen3.8-Omni-Flash-Realtime, and keeps Qwen3.5-Omni as its pick for spoken replies on the standard API.

How does Qwen3.5-Omni compare with Qwen3.8-Omni-Flash?

Qwen3.8-Omni-Flash is newer, cheaper and in more regions, but its standard API answers in text only; Qwen3.5-Omni is the one Alibaba keeps for spoken replies.

Output

Qwen3.8-Omni-Flash writes text; its separate Realtime version speaks. Qwen3.5-Omni speaks on both its standard and Realtime APIs.

Price

Qwen3.8-Omni-Flash costs $0.15 in and $0.47 out per million tokens in Singapore, against $0.40 and $2.20 for Qwen3.5-Omni-Flash.

Context

1M tokens with up to 131,072 tokens of output, against 256K and 65,536 for Qwen3.5-Omni.

Regions and thinking

Qwen3.8-Omni-Flash runs in all six Model Studio regions and can think before answering; Qwen3.5-Omni runs in two and has no thinking mode.

Who is Qwen3.5-Omni best for?

Qwen3.5-Omni is best for developers building voice assistants, transcription or video-understanding apps that need one model to take in text, images, audio and video and reply in text or speech.

• Long recordings and video

  One API request takes audio files up to 3 hours or video files up to 1 hour, inside a 256K-token context.

• Multilingual voice products

  It recognises speech in 113 languages and dialects and speaks 36, so one model covers many markets.

• Spoken replies from files

  Alibaba still names it the model to use when a standard, non-live API call has to answer in a voice.

• Detailed captions and transcripts

  Plus writes structured, time-stamped descriptions of audio and video that name the speakers and describe the sounds.

What's included with Qwen3.5-Omni?

Spoken answers in 55 voices, voice cloning and voice control, interruption handling for live calls, web search and tool calls, and mixed inputs in one request.

Speech output

Answers in text or in one of 55 voices, 47 of them multilingual and 8 in dialects.

Voice cloning

Makes a custom voice from 10 to 20 seconds of clear speech, for Plus, Flash and both Realtime models.

Voice control

Follows instructions to change its volume, speaking speed and emotion.

Semantic interruption

The Realtime versions tell a real interruption apart from a listener's 'mm-hm' or background noise.

Web search and tool calls

Every version can search the web; function calls work on the Realtime versions, and on the standard API only in Beijing.

Mixed inputs in one request

Any mix of text, images, audio and video in a single call, with files up to 2 GB by URL.

How much does Qwen3.5-Omni cost?

Qwen3.5-Omni is billed per token on Alibaba Cloud Model Studio, from $0.40 per million input tokens and $2.20 per million output tokens for Qwen3.5-Omni-Flash up to $2.10 and $12.40 for Qwen3.5-Omni-Plus-Realtime, with audio in and out priced higher.

These are pay-as-you-go prices for the Singapore (International) region; Model Studio does not offer Qwen3.5-Omni in its US (Virginia) region. In the China (Beijing) region, Plus costs $0.96 in and $5.50 out and Flash $0.30 in and $1.83 out per million tokens for text, image or video in and text out. Each model, Realtime ones included, comes with 1 million free tokens in Singapore, valid for 90 days from Model Studio activation or the model's release, whichever is later. The Light size named in the launch post has no API price. Full pricing on alibabacloud.com. We checked these prices on 28 Sep 2026.

How do readers rate Qwen3.5-Omni?

No one has rated Qwen3.5-Omni yet. Be the first.

Qwen3.5-Omni pros and cons

It leads Gemini 3.1 Pro on most audio tests in Qwen's tables and speaks 36 languages and dialects, but the weights are closed, it runs in only two regions, and newer Qwen3.8 models cost less.

What are the best Qwen3.5-Omni alternatives?

Four tools on Recaply do Qwen3.5-Omni's job. Step-Audio-R1 is open source, so you can run it on your own server.

GPT-Realtime-2Paid

Choose it when you want OpenAI's live voice agent with adjustable reasoning effort; its audio input costs $32 per million tokens, about twice Qwen3.5-Omni-Plus-Realtime's $16.50.

Gemini 3.8 LiveFree

Choose it when you want a free tier to build on and cheaper live audio: $3 in and $12 out per million audio tokens, against $4.50 and $17.70 on Qwen3.5-Omni-Flash-Realtime.

Nemotron 3 Nano OmniFree

Choose it when you must run audio and video understanding on your own GPUs; its weights are open, but it answers only in English text.

Step-Audio-R1Open Source

Choose it when you want an Apache-licensed model that reasons over speech, sounds and music before answering.

Questions about Qwen3.5-Omni