Tongyi Qianwen

Tongyi Qianwen

7/12/2025

Overview

Alibaba's multimodal AI assistant

Details

Overview

Tongyi Qianwen (通义千问, now branded 千问/Qianwen) is Alibaba's self-developed, multimodal AI assistant and large-model family, built by Alibaba's Tongyi Lab. First launched in April 2023, the series iterated to Qwen3.7 in May 2026 and Qwen3.8-MAX in August 2026 — and it is one of the first domestically registered (网信办) LLMs in China. The assistant grew explosively: after renaming to "Qianwen" in November 2025, monthly active users hit 100M in January 2026 and 165M on mobile by March 2026 (QuestMobile, +4241% YoY), with all-ends (App + Web + PC) MAU surpassing 300M in February 2026.

Key Features

  • AI errands & life services (the core moat): unlike rivals, Qianwen completes real transactions in-chat — ordering takeout, hailing rides, booking movie tickets and hotels end-to-end via Alibaba-owned infrastructure (Taobao Flash Purchase, Alipay AI Pay, Fliggy, Amap, Damai, Hema). AI phone calls dial merchants for you with real-time emotion/intent recognition (50+ emotions in ~100ms) and return a written summary plus recording.
  • Multimodal model family (Qwen3.7 series): Qwen3.7-Max (flagship, ~100M-token context, near GPT-4/Claude Opus 4.5-level reasoning), Qwen3.7-Plus (balanced, 1M context, ~1/3 the cost of Max), Qwen3.7-Flash (low-latency lightweight), plus Qwen-VL (vision/OCR), Qwen-Audio (speech) and Wanxiang 2.7 (image/video generation, 20B params). Qwen3.8-Max-Preview (2.4T-param MoE, Tongyi's first trillion-parameter multimodal model) launched mid-2026.
  • 15 quick features: Task Assistant (multi-step planning agent that parallel-processes up to 100 files on web, ~8–10 min per report, with dual-verification), homework photo QA and grading, AI image/video generation, AI writing, PPT creation, real-time recording with system-audio capture, AI photo retouching, translation, deep research and document reading.
  • HappyHorse video generation: a 15B-param model with native audio+video joint generation, 7-language lip sync and ~38s per clip on a single H100 — it topped Artificial Analysis's text-to-video and image-to-video charts in April 2026 and is free for regular users.
  • Agent & developer ecosystem: OpenAI-compatible APIs and agent development on Alibaba Cloud Bailian, private deployment, and vertical products like Qwen-Coder and Omni models for devices, toys and safety screening.
  • Free for everyone (Aug 2026): five new features — Thinking Research, Scheduled Tasks, Office Assistant, Agent Plaza and Voice Calls — powered by Qwen3.8-MAX and free for all users.

Pricing

  • Free: core chat, short-text generation and simple Q&A are free forever with no ads and no quotas; regular users also get free AI image/video generation (e.g. HappyHorse) and daily creation allowances.
  • Video generation packs (in-app): ~¥1 for 2, ¥26 for 10, ¥116 for 50, ¥228 for 100, ¥968 for 500 generations.
  • Personal Pro (Token Plan): ¥59/month or ¥599/year — 20M tokens/month, access to all models, multimodal generation, agent basics and ad-free.
  • Team plans (per seat/month): Standard ¥198 (25,000 credits), Advanced ¥698 (100,000 credits), Premium ¥1,398 (250,000 credits) — team management, permissions, unified billing, enterprise agents and dedicated support.
  • Enterprise: custom annual contracts — unlimited tokens, private/VPC deployment with data staying in-house, model fine-tuning, dedicated manager and 24/7 support.
  • API (per 1M tokens): Qwen3.7-Max ¥2 input / ¥4 output; Qwen3.7-Plus ¥1 / ¥2 (Max often discounted to 50% during promotions). New Bailian users get 70M free tokens (90 days); verified students get an extra 30M.

Best For

Chinese consumers and students who want a free, all-in-one AI assistant that can actually complete daily tasks (ordering, bookings, homework, video creation), plus developers and enterprises that need Alibaba-grade multimodal models with strong Chinese understanding, OpenAI-compatible APIs, or fully private deployment.

FAQ

Is Tongyi Qianwen free?

Yes — core chat, writing, translation, homework help, image/video generation basics and the five new features (Thinking Research, Office Assistant, etc.) are free. Paid plans add higher token allowances, more video generations and team/enterprise features.

What makes Qianwen different from Doubao or DeepSeek?

Its Alibaba-ecosystem "AI errands" loop: it can place real orders (takeout, rides, tickets) with in-chat Alipay payment — something no competitor's infrastructure can easily replicate.

What models does it run?

The Qwen3.7 family (Max/Plus/Flash) plus vision (Qwen-VL), audio (Qwen-Audio) and image/video generation (Wanxiang 2.7, HappyHorse), with Qwen3.8-MAX now free to all users.

How do developers access it?

Via Alibaba Cloud Bailian with OpenAI-compatible APIs (per-token pricing from ¥1/M tokens) or Token Plan subscriptions; enterprise offers private deployment and fine-tuning.