Alibaba's Qwen team released Qwen3-TTS last week. It’s an open-source text-to-speech system that runs on consumer hardware while matching the quality of expensive commercial APIs. The timing matters because voice AI has been locked behind paywalls for years, forcing developers to choose between quality and cost.
The release comes with Apache 2.0 licensing, which means anyone can use it commercially without fees or restrictions. This represents a genuine shift in how AI voice technology gets distributed and who gets to build with it.
What can it do?
Qwen3-TTS, an open-source text-to-speech model family trained on over 5 million hours of speech data across 10 languages.
The system can clone voices from just 3 seconds of audio and create entirely new voices from natural language descriptions like "deep male voice with British accent."
The smallest model runs on approximately 2GB of VRAM, meaning it works on consumer GPUs while the larger 1.7 billion parameter version still fits on standard gaming hardware.
The dual-track architecture achieves latency as low as 97 milliseconds for first-packet delivery, enabling real-time streaming conversations.
People have already started comparing this to OpenAI's Whisper release, calling it a potential watershed moment for open-source voice synthesis.
You can check out the release blog here and try out a demo on Hugging Face.
Why does this matter?
The best voice AI systems from companies like ElevenLabs and OpenAI charge per character or per minute, creating barriers for developers building voice agents, educational tools, or experimental projects where budget constraints matter.
We’re subscribers to Eleven Labs, and while the superior UX of the application still makes a subscription worthwhile, an open source alternative could soon be published that obviates the need for paying a $20+ a month subscription. And this could start happening more and more.
Our take
The pattern repeats. Almost exactly a year ago DeepSeek caused shockwaves in global markets as an open source Large Language Model alternative to expensive US models. Strange censorship aside, a competitive chatbot was made available for free. Likewise Qwen’s chatbot was available too.
Late last year Tim Dettmers published a blog titled Why AGI Will Not Happen. An interesting point is that in the US, literally trillions of dollars are being spent on infrastructure for powerful AI, and potentially superintelligence. China’s strategy, on the other hand, They believe model capabilities do not matter as much as application.
Open sourcing the technology is part of a diffusion model, that enables economic productivity gains through ‘good enough’ AI.
Another big thing… Claude’s new constitution
Surely timed for the PR busting appearance of Anthropic CEO Dario Amodei at the World Economic Forum last week, Claude got a new constitution. It’s some 20,000 words long - but it raises some interesting points around AI safety and even consciousness.
Grab the URL and use one of these summariser prompts to give yourself a lowdown.
This was ‘The Thing That Matters in AI’ - a new edition of Absolutely Agentic that we will be sending every Monday at 1pm UK time during January.
