Cryptonews

Tiny Tether Model Outperforms Google Counterpart Despite Being 87% More Compact

Source
CryptoNewsTrend
Published
Tiny Tether Model Outperforms Google Counterpart Despite Being 87% More Compact

Tether, the company behind the USDT stablecoin, released two artificial intelligence (AI) models for medical use in May 2026. The models are part of Tether's QVAC AI group. QVAC is a decentralised, local-first AI framework that runs directly on a user's device. The models are named MedPsy-1.7B and MedPsy-4B. The numbers refer to the size of each model, measured in billions of parameters. Parameters are the internal values an AI model uses to process and generate responses.

Models run without internet connectionBoth MedPsy models operate without sending data to a remote server. This is called on-device or local inference. QVAC's SDK launch materials state that apps built on QVAC continue to function when a network is unavailable.

"if the internet goes down, the AI keeps working.", May 2026. — QVAC / Tether, SDK Launch Materials, Tether Ltd. 

MedPsy-1.7B outscores larger Google modelQVAC measured both models across seven closed-ended medical benchmarks. A benchmark is a standardised test used to compare AI performance. MedPsy-1.7B scored 62.62 on average. Google's comparable model, MedGemma-1.5-4B-it, scored 51.20. MedPsy-1.7B contains fewer than half the parameters of the Google model it outscored. These figures come from QVAC's own Hugging Face technical report published 06 May 2026. Independent laboratory replication had not been completed as of that date.

MedPsy-4B edges past 27-billion-parameter rivalThe larger MedPsy-4B model scored 70.54 on the same benchmark average. Google's MedGemma-27B-text-it scored 69.95. MedPsy-4B contains approximately 4 billion parameters. MedGemma-27B contains approximately 27 billion parameters, making it nearly seven times larger. QVAC also tested both models on HealthBench Hard, a separate medical reasoning benchmark evaluated with CompassJudger. MedPsy-4B scored 58.00 on that test. MedGemma-27B-text-it scored 42.00 on the same test.

Compressed file sizes fit on mobile devicesQVAC released the models in GGUF format, a file type used to run AI models on consumer hardware. A compressed version called Q4_K_M reduces the MedPsy-4B file to 2.72 GB. QVAC's Hugging Face model card states this compression cuts file size by 69% while reducing benchmark performance by only 0.81 points on average. The MedPsy-4B model card, published 06 May 2026, confirms the 2.72 GB figure directly.

"Q4_K_M is the sweet spot: only -0.81 pts (-1.12% relative) AVG Score loss for a 69% size reduction (2.72 GB), comfortably fitting on high-end mobile and laptop devices.", 06 May 2026. — QVAC Research Team, Tether AI Group 

Tether reports $1.04 billion Q1 2026 profitTether published its Q1 2026 financial attestation from 30 April 2026. An attestation is a financial review by an external auditor, in this case the firm BDO. The report recorded $1.04 billion in net profit for the quarter. Tether held $8.23 billion in reserve buffer above its liabilities. Total token-related liabilities reached $183.5 billion. The company held approximately $141.2 billion in direct and indirect exposure to United States Treasury bills, which are short-term government debt instruments. Both MedPsy models operate without sending data to a remote server. This is called on-device or local inference. QVAC's SDK launch materials state that apps built on QVAC continue to function when a network is unavailable.

"if the internet goes down, the AI keeps working.", May 2026. — QVAC / Tether, SDK Launch Materials, Tether Ltd. 

MedPsy-1.7B outscores larger Google modelQVAC measured both models across seven closed-ended medical benchmarks. A benchmark is a standardised test used to compare AI performance. MedPsy-1.7B scored 62.62 on average. Google's comparable model, MedGemma-1.5-4B-it, scored 51.20. MedPsy-1.7B contains fewer than half the parameters of the Google model it outscored. These figures come from QVAC's own Hugging Face technical report published 06 May 2026. Independent laboratory replication had not been completed as of that date.

MedPsy-4B edges past 27-billion-parameter rivalThe larger MedPsy-4B model scored 70.54 on the same benchmark average. Google's MedGemma-27B-text-it scored 69.95. MedPsy-4B contains approximately 4 billion parameters. MedGemma-27B contains approximately 27 billion parameters, making it nearly seven times larger. QVAC also tested both models on HealthBench Hard, a separate medical reasoning benchmark evaluated with CompassJudger. MedPsy-4B scored 58.00 on that test. MedGemma-27B-text-it scored 42.00 on the same test.

Compressed file sizes fit on mobile devicesQVAC released the models in GGUF format, a file type used to run AI models on consumer hardware. A compressed version called Q4_K_M reduces the MedPsy-4B file to 2.72 GB. QVAC's Hugging Face model card states this compression cuts file size by 69% while reducing benchmark performance by only 0.81 points on average. The MedPsy-4B model card, published 06 May 2026, confirms the 2.72 GB figure directly.

"Q4_K_M is the sweet spot: only -0.81 pts (-1.12% relative) AVG Score loss for a 69% size reduction (2.72 GB), comfortably fitting on high-end mobile and laptop devices.", 06 May 2026. — QVAC Research Team, Tether AI Group 

Tether reports $1.04 billion Q1 2026 profitTether published its Q1 2026 financial attestation from 30 April 2026. An attestation is a financial review by an external auditor, in this case the firm BDO. The report recorded $1.04 billion in net profit for the quarter. Tether held $8.23 billion in reserve buffer above its liabilities. Total token-related liabilities reached $183.5 billion. The company held approximately $141.2 billion in direct and indirect exposure to United States Treasury bills, which are short-term government debt instruments. "if the internet goes down, the AI keeps working.", May 2026. — QVAC / Tether, SDK Launch Materials, Tether Ltd. 

MedPsy-1.7B outscores larger Google modelQVAC measured both models across seven closed-ended medical benchmarks. A benchmark is a standardised test used to compare AI performance. MedPsy-1.7B scored 62.62 on average. Google's comparable model, MedGemma-1.5-4B-it, scored 51.20. MedPsy-1.7B contains fewer than half the parameters of the Google model it outscored. These figures come from QVAC's own Hugging Face technical report published 06 May 2026. Independent laboratory replication had not been completed as of that date.

MedPsy-4B edges past 27-billion-parameter rivalThe larger MedPsy-4B model scored 70.54 on the same benchmark average. Google's MedGemma-27B-text-it scored 69.95. MedPsy-4B contains approximately 4 billion parameters. MedGemma-27B contains approximately 27 billion parameters, making it nearly seven times larger. QVAC also tested both models on HealthBench Hard, a separate medical reasoning benchmark evaluated with CompassJudger. MedPsy-4B scored 58.00 on that test. MedGemma-27B-text-it scored 42.00 on the same test.

Compressed file sizes fit on mobile devicesQVAC released the models in GGUF format, a file type used to run AI models on consumer hardware. A compressed version called Q4_K_M reduces the MedPsy-4B file to 2.72 GB. QVAC's Hugging Face model card states this compression cuts file size by 69% while reducing benchmark performance by only 0.81 points on average. The MedPsy-4B model card, published 06 May 2026, confirms the 2.72 GB figure directly.

"Q4_K_M is the sweet spot: only -0.81 pts (-1.12% relative) AVG Score loss for a 69% size reduction (2.72 GB), comfortably fitting on high-end mobile and laptop devices.", 06 May 2026. — QVAC Research Team, Tether AI Group 

Tether reports $1.04 billion Q1 2026 profitTether published its Q1 2026 financial attestation from 30 April 2026. An attestation is a financial review by an external auditor, in this case the firm BDO. The report recorded $1.04 billion in net profit for the quarter. Tether held $8.23 billion in reserve buffer above its liabilities. Total token-related liabilities reached $183.5 billion. The company held approximately $141.2 billion in direct and indirect exposure to United States Treasury bills, which are short-term government debt instruments. MedPsy-1.7B outscores larger Google modelQVAC measured both models across seven closed-ended medical benchmarks. A benchmark is a standardised test used to compare AI performance. MedPsy-1.7B scored 62.62 on average. Google's comparable model, MedGemma-1.5-4B-it, scored 51.20. MedPsy-1.7B contains fewer than half the parameters of the Google model it outscored. These figures come from QVAC's own Hugging Face technical report published 06 May 2026. Independent laboratory replication had not been completed as of that date.

MedPsy-4B edges past 27-billion-parameter rivalThe larger MedPsy-4B model scored 70.54 on the same benchmark average. Google's MedGemma-27B-text-it scored 69.95. MedPsy-4B contains approximately 4 billion parameters. MedGemma-27B contains approximately 27 billion parameters, making it nearly seven times larger. QVAC also tested both models on HealthBench Hard, a separate medical reasoning benchmark evaluated with CompassJudger. MedPsy-4B scored 58.00 on that test. MedGemma-27B-text-it scored 42.00 on the same test.

Compressed file sizes fit on mobile devicesQVAC released the models in GGUF format, a file type used to run AI models on consumer hardware. A compressed version called Q4_K_M reduces the MedPsy-4B file to 2.72 GB. QVAC's Hugging Face model card states this compression cuts file size by 69% while reducing benchmark performance by only 0.81 points on average. The MedPsy-4B model card, published 06 May 2026, confirms the 2.72 GB figure directly.

"Q4_K_M is the sweet spot: only -0.81 pts (-1.12% relative) AVG Score loss for a 69% size reduction (2.72 GB), comfortably fitting on high-end mobile and laptop devices.", 06 May 2026. — QVAC Research Team, Tether AI Group 

Tether reports $1.04 billion Q1 2026 profitTether published its Q1 2026 financial attestation from 30 April 2026. An attestation is a financial review by an external auditor, in this case the firm BDO. The report recorded $1.04 billion in net profit for the quarter. Tether held $8.23 billion in reserve buffer above its liabilities. Total token-related liabilities reached $183.5 billion. The company held approximately $141.2 billion in direct and indirect exposure to United States Treasury bills, which are short-term government debt instruments. QVAC measured both models across seven closed-ended medical benchmarks. A benchmark is a standardised test used to compare AI performance. MedPsy-1.7B scored 62.62 on average. Google's comparable model, MedGemma-1.5-4B-it, scored 51.20. MedPsy-1.7B contains fewer than half the parameters of the Google model it outscored. These figures come from QVAC's own Hugging Face technical report published 06 May 2026. Independent laboratory replication had not been completed as of that date.

MedPsy-4B edges past 27-billion-parameter rivalThe larger MedPsy-4B model scored 70.54 on the same benchmark average. Google's MedGemma-27B-text-it scored 69.95. MedPsy-4B contains approximately 4 billion parameters. MedGemma-27B contains approximately 27 billion parameters, making it nearly seven times larger. QVAC also tested both models on HealthBench Hard, a separate medical reasoning benchmark evaluated with CompassJudger. MedPsy-4B scored 58.00 on that test. MedGemma-27B-text-it scored 42.00 on the same test.

Compressed file sizes fit on mobile devicesQVAC released the models in GGUF format, a file type used to run AI models on consumer hardware. A compressed version called Q4_K_M reduces the MedPsy-4B file to 2.72 GB. QVAC's Hugging Face model card states this compression cuts file size by 69% while reducing benchmark performance by only 0.81 points on average. The MedPsy-4B model card, published 06 May 2026, confirms the 2.72 GB figure directly.

"Q4_K_M is the sweet spot: only -0.81 pts (-1.12% relative) AVG Score loss for a 69% size reduction (2.72 GB), comfortably fitting on high-end mobile and laptop devices.", 06 May 2026. — QVAC Research Team, Tether AI Group 

Tether reports $1.04 billion Q1 2026 profitTether published its Q1 2026 financial attestation from 30 April 2026. An attestation is a financial review by an external auditor, in this case the firm BDO. The report recorded $1.04 billion in net profit for the quarter. Tether held $8.23 billion in reserve buffer above its liabilities. Total token-related liabilities reached $183.5 billion. The company held approximately $141.2 billion in direct and indirect exposure to United States Treasury bills, which are short-term government debt instruments. The larger MedPsy-4B model scored 70.54 on the same benchmark average. Google's MedGemma-27B-text-it scored 69.95. MedPsy-4B contains approximately 4 billion parameters. MedGemma-27B contains approximately 27 billion parameters, making it nearly seven times larger. QVAC also tested both models on HealthBench Hard, a separate medical reasoning benchmark evaluated with CompassJudger. MedPsy-4B scored 58.00 on that test. MedGemma-27B-text-it scored 42.00 on the same test.

Compressed file sizes fit on mobile devicesQVAC released the models in GGUF format, a file type used to run AI models on consumer hardware. A compressed version called Q4_K_M reduces the MedPsy-4B file to 2.72 GB. QVAC's Hugging Face model card states this compression cuts file size by 69% while reducing benchmark performance by only 0.81 points on average. The MedPsy-4B model card, published 06 May 2026, confirms the 2.72 GB figure directly.

"Q4_K_M is the sweet spot: only -0.81 pts (-1.12% relative) AVG Score loss for a 69% size reduction (2.72 GB), comfortably fitting on high-end mobile and laptop devices.", 06 May 2026. — QVAC Research Team, Tether AI Group 

Tether reports $1.04 billion Q1 2026 profitTether published its Q1 2026 financial attestation from 30 April 2026. An attestation is a financial review by an external auditor, in this case the firm BDO. The report recorded $1.04 billion in net profit for the quarter. Tether held $8.23 billion in reserve buffer above its liabilities. Total token-related liabilities reached $183.5 billion. The company held approximately $141.2 billion in direct and indirect exposure to United States Treasury bills, which are short-term government debt instruments. QVAC released the models in GGUF format, a file type used to run AI models on consumer hardware. A compressed version called Q4_K_M reduces the MedPsy-4B file to 2.72 GB. QVAC's Hugging Face model card states this compression cuts file size by 69% while reducing benchmark performance by only 0.81 points on average. The MedPsy-4B model card, published 06 May 2026, confirms the 2.72 GB figure directly.

"Q4_K_M is the sweet spot: only -0.81 pts (-1.12% relative) AVG Score loss for a 69% size reduction (2.72 GB), comfortably fitting on high-end mobile and laptop devices.", 06 May 2026. — QVAC Research Team, Tether AI Group 

Tether reports $1.04 billion Q1 2026 profitTether published its Q1 2026 financial attestation from 30 April 2026. An attestation is a financial review by an external auditor, in this case the firm BDO. The report recorded $1.04 billion in net profit for the quarter. Tether held $8.23 billion in reserve buffer above its liabilities. Total token-related liabilities reached $183.5 billion. The company held approximately $141.2 billion in direct and indirect exposure to United States Treasury bills, which are short-term government debt instruments. "Q4_K_M is the sweet spot: only -0.81 pts (-1.12% relative) AVG Score loss for a 69% size reduction (2.72 GB), comfortably fitting on high-end mobile and laptop devices.", 06 May 2026. — QVAC Research Team, Tether AI Group 

Tether reports $1.04 billion Q1 2026 profitTether published its Q1 2026 financial attestation from 30 April 2026. An attestation is a financial review by an external auditor, in this case the firm BDO. The report recorded $1.04 billion in net profit for the quarter. Tether held $8.23 billion in reserve buffer above its liabilities. Total token-related liabilities reached $183.5 billion. The company held approximately $141.2 billion in direct and indirect exposure to United States Treasury bills, which are short-term government debt instruments. Tether reports $1.04 billion Q1 2026 profitTether published its Q1 2026 financial attestation from 30 April 2026. An attestation is a financial review by an external auditor, in this case the firm BDO. The report recorded $1.04 billion in net profit for the quarter. Tether held $8.23 billion in reserve buffer above its liabilities. Total token-related liabilities reached $183.5 billion. The company held approximately $141.2 billion in direct and indirect exposure to United States Treasury bills, which are short-term government debt instruments. Tether published its Q1 2026 financial attestation from 30 April 2026. An attestation is a financial review by an external auditor, in this case the firm BDO. The report recorded $1.04 billion in net profit for the quarter. Tether held $8.23 billion in reserve buffer above its liabilities. Total token-related liabilities reached $183.5 billion. The company held approximately $141.2 billion in direct and indirect exposure to United States Treasury bills, which are short-term government debt instruments. Cryptocurrencies are highly volatile and involve significant risk. You may lose part or all of your investment. All information on Coinpaprika is provided for informational purposes only and does not constitute financial or investment advice. Always conduct your own research (DYOR) and consult a qualified financial advisor before making investment decisions. Coinpaprika is not liable for any losses resulting from the use of this information.

Tiny Tether Model Outperforms Google Counterpart Despite Being 87% More Compact