Tether’s compact models run AI tasks on-device without cloud
Tether’s lightweight AI models challenge the trend toward ever-larger models by delivering specialized capabilities on consumer hardware. Their latest suite includes models like MedPsy, TranslatePsy, and VisionPsy, each optimized for tasks such as medical reasoning, translation, and visual processing. These models require far less compute than frontier systems, with some running on basic…
Key points
- Tether’s models run on-device without continuous cloud access, unlike most AI systems
- Genesis III dataset contains 191 billion tokens across 19 STEM domains for specialized training
- QVAC SDK includes TurboQuant (80% KV cache reduction) and TurboVec for efficient inference
The company’s QVAC research team achieves efficiency through data curation and compression techniques. Their Genesis III dataset, for example, includes 191 billion tokens across 19 STEM domains, designed as structured educational materials rather than generic web-scraped data. Quantization—replacing high-precision weights with low-bit alternatives—further reduces memory usage. The QVAC SDK also includes TurboQuant (80% KV cache reduction) and TurboVec, enabling clustered inference for workloads exceeding single-device limits. The result: cloud-dependent AI tasks now run locally, addressing scalability limits and data center costs.
No signal, no problem: Smaller models bring more AI capabilities on-device
computerworld.com · 22 October 2026
Loading the full article…
This text was published by computerworld.com. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Researchers combine GPT-Neo with Random Forest to predict Agrobacterium promoters at 97% accuracy · 1 src
- OpenAI model allegedly disproves Exponential Time Hypothesis · 6 src
- SLICE-Mem cuts LLM memory costs by 99.5% in new framework · 1 src
- Researchers propose RSD metric to measure LLM agent consistency · 1 src
- Researchers propose tracequery to cut diffusion LLM inference costs · 1 src
Comments
via GitHub Discussions