Tether AI Team Launches QVAC MedPsy Series Local Medical Language Models
Tether (USDT issuer) AI research team today released the QVAC MedPsy series of medical language models, designed for low-power terminals such as smartphones and wearable devices.
This series can operate without cloud servers, achieving performance far exceeding parameter scale through an efficient architecture:
1.7B parameter version: Average score of 62.62 on seven closed medical benchmarks, surpassing Google MedGemma-4B by 11.42 points, and outperforming the nearly 16 times larger MedGemma-27B in real clinical scenarios like HealthBench Hard.
4B parameter version: Score as high as 70.54, while significantly reducing inference token consumption (up to 3.2 times).
The models are released in quantized GGUF format (1.7B version is only about 1.2GB), making them extremely suitable for deployment on mobile and edge devices.
Source: Public Information
ABAB AI Insight
Tether AI team's release focuses on "localization + high efficiency," achieving breakthroughs in the medical field with small models through phased medical post-training (supervised fine-tuning + clinical reasoning data + reinforcement learning), directly challenging the traditional assumption that "large models = strong performance." Tether CEO Paolo Ardoino emphasized that this model can process sensitive medical data locally in hospitals or on devices without uploading to the cloud, significantly reducing latency, costs, and risks of privacy breaches.
On the capital path, Tether is directing the stable cash flow generated by its stablecoin business towards AI infrastructure, promoting the global adoption of medical AI, especially in underdeveloped regions, laying the foundation for building a localized medical AI ecosystem in emerging markets.
Similar to the layouts of Apple Intelligence and Qualcomm's edge AI chips, Tether is currently in the early stages of stablecoin giants expanding from financial services to practical applications in AI + healthcare.
Essentially a technological replacement: QVAC MedPsy replaces cloud-dependent large model medical AI with efficient small models, restructuring capital from high-cost centralized cloud inference to local privacy protection + low-latency device operation. Mechanically, this is achieved through quantization compression and targeted post-training, allowing medical AI to truly achieve "offline availability," accelerating the global healthcare infrastructure's shift from cloud dependency to distributed local deployment.
ABAB News · Cognitive Law
The best medical AI is not the one with the most parameters, but the one that can run safely on a phone.
When the 1.7B model beats the 27B large model, the era of "bigger is better" officially ends.
Truly inclusive technology is one that allows the places most in need of medical AI to no longer rely on expensive cloud services.