Dec 12, 2025 | The Tongyi Weekly: Your weekly dose of cutting-edge AI from Tongyi Lab
Published: (December 12, 2025 at 12:59 AM EST)
3 min read
Source: Dev.to
Source: Dev.to
Qwen3‑Omni‑Flash (2025‑12‑01) – Smarter, More Human
- Enhanced multi‑turn video/audio understanding – conversations flow naturally.
- Customize your AI’s personality through system prompts (think role‑play scenarios!).
- Smarter language handling + rock‑solid support: 119 text languages | 19 speech.
- Voices indistinguishable from humans.
Try it now:
Qwen Chat – click the VoiceChat and VideoChat button (bottom‑right).
Qwen3‑TTS (2025‑11‑27) – Voices That Feel Real
- More Personalities: Over 49 high‑quality voices, from cute and playful to wise and stern.
- Global Reach: Supports 10 languages (zh, en, de, it, pt, es, ja, ko, fr, ru) and authentic dialects (Minnan, Wu, Cantonese, Sichuan, Beijing, Nanjing, Tianjin, Shaanxi).
- Insanely Natural: Rhythm and speed adapt just like a real person.
Try it now:
Qwen Chat – click Response → Read aloud.
Qwen Code v0.2.2 → v0.3.0: Stream JSON + Global Ready
Stream JSON Support
--output-format stream-jsonfor streaming output.--input-format stream-jsonfor structured input.- 3‑tier adapter architecture + complete session management.
- Endless possibilities for SDK integration, automation tools, CI/CD pipelines.
Full Internationalization
- Built‑in EN/CN interface + custom language‑pack extensions.
/language ui zh-CN– one‑click UI switching./language output Chinese– set AI output language.- Global developers welcome to contribute local language packs.
Security & Stability Leap Forward
Qwen Learn Mode — Your Personal AI Learning Tutor
- Guides you through Socratic‑style dialogue instead of just giving answers.
- Adapts to your current level, keeping you in an optimal learning zone.
- Builds mental scaffolds so you can handle complex logic without feeling overwhelmed.
✨ Try Learn Mode
SAPO: A Smoother Path to RL Training
- Smooth trust‑region behavior → no abrupt gradient drop.
- Sequence‑level coherence → aligned sequence‑level behavior.
- Token‑level adaptivity → preserves useful gradients & boosts sample efficiency.
- Asymmetric temperatures → significantly improved stability, especially in MoE models.
What this means in practice:
- Longer stable RL runs.
- Higher Pass@1.
- Stronger performance on Qwen3‑VL across math, coding & multimodal tasks.
Model Milestone: Z‑Image‑Turbo
- #1 Open Source Model – top 10 overall, the only open model on the list.
- High‑fidelity outputs, $5/1k pricing, fully open source.
- Accessible, affordable, and community‑driven generative AI.
Qwen3‑4B: The #1 Base Model for Fine‑Tuning
XiYan‑SQL: #1 on All Open BIRD‑CRITIC Leaderboards
- Not just text → SQL: diagnoses and fixes failing queries.
- Handles complex operations (INSERT / UPDATE / DELETE) across messy, multi‑dialect databases.
- Remains robust on unseen, out‑of‑distribution databases.
What this means in practice:
- More reliable SQL debugging in real, production‑style environments.
- Stronger robustness for messy and evolving data stacks.
Community Celebration: WanMuse+ “Heartbeat” Winners Announced
🎉 Congratulations to all finalists and winners!
Light Migration LoRA: Qwen‑Edit‑2509‑Light‑Migration
Upscale LoRA: Qwen‑Image‑Edit‑2509‑Upscale2K
Subscribe
Never miss a release:
Subscribe to The Tongyi Weekly
About Tongyi Lab
Tongyi Lab is a research institution under Alibaba Group dedicated to artificial intelligence and foundation models. Its work spans large language models (LLMs), multimodal understanding and generation, visual AIGC, speech technologies, and more.