Tether launches $5.38 million AI fund for USDT and BTC rewards

Tether’s million-dollar bet on local AI

Tether has allocated significant resources to its initiative supporting developer grants for local AI and payment infrastructure projects, without imposing any upper funding limit. Previously, Tether’s AI research team released medical language models capable of outperforming much larger Google models—while running even on ordinary smartphones. Under the new program, developers will receive between $1,500 and $4,000 per project, paid in either USDT or Bitcoin. Funds are disbursed once specific technical milestones are achieved.

With this move, Tether is following in the footsteps of Ethereum co-founder Vitalik Buterin, who recently announced his own switch from cloud-based AI to device-powered systems. Security has become a growing concern, especially given that emerging “agent” technologies can open the door to data breaches. For example, a tool named “OpenClaw” can enable AI agents to access crucial computer settings, download malware unseen by users, and quietly transmit data to remote servers.

Paolo Ardoino has stated, “If you are able to build a fully local system that creates direct value and doesn’t depend on external providers, we are ready to support you.”

Within the new grant program, Tether plans to prioritize the creation of core libraries, comprehensive documentation, open-source stack development, and research into decentralized AI models. A key focus is the Wallet Development Kit (WDK), which empowers developers to embed self-custody wallets directly into apps—enabling users to manage their accounts and transactions without third-party involvement.

Big results from compact AI models

Tether’s AI research division recently introduced QVAC MedPsy, a set of medical language models designed to run on smartphones and wearables without needing internet access. Trials have demonstrated that these streamlined models can sometimes outperform far larger, cloud-based AI systems.

The QVAC MedPsy-1.7B model, with 1.7 billion parameters, scored 62.62 points across tests in seven medical fields—beating Google’s MedGemma-1.5-4B-it model by 11.42 points. Meanwhile, the more advanced QVAC MedPsy-4B, with 4 billion parameters, achieved 70.54 points, even surpassing the much larger Google MedGemma-27B-text-it model with its approximately 27 billion parameters.

On the HealthBench Hard test, designed to mimic real-world clinical scenarios, the MedPsy-4B model scored 58.00 points—substantially outperforming Google’s MedGemma-27B, which earned just 42.00. Tether’s models also stand out for their efficiency, requiring up to three times fewer tokens per message, resulting in faster response times and reduced computational needs compared to similar solutions.

Industry leaders like Tether and Vitalik Buterin see powerful potential in compact, locally-run AI technologies, arguing that they offer a compelling alternative to cloud-based systems—both in terms of efficiency and user data privacy.

Tether expands grants and education funding

Tether’s commitment to supporting developers is not new; in prior years, the company granted $100,000 for two consecutive years to the BTC Pay Server Foundation and made a $250,000 Bitcoin donation to the OpenSats initiative for Bitcoin development. The company has also awarded scholarships to over 500 students and now pledges to allocate up to $5.38 million for new grants and educational programs by 2030.