Tencent's Hunyuan team has released an extreme quantization version of the Hy4 preview model, compressing the original 1.5TB weights to about 214GB in GGUF format. Using a layer-wise quantization strategy, the average bit-width is 2.38 bpw, with performance loss of only 0.2-1.6 points on benchmarks. The team also tested heterogeneous joint inference with prima.cpp, achieving 1.02 token/s on a setup combining an RTX 4090 laptop and a quad-RTX A4000 server, 6x faster than laptop-only offloading.

Tether’s QVAC team unveils QVAC MedPsy, a medical language model that runs offline on smartphones, solving privacy concerns. Its 1.7B parameter version outperforms Google’s MedGemma 27B, which is 16 times larger. CEO Paolo Ardoino highlights efficiency and local deployment.

Moonshot AI this week unveiled Kimi K3, a 2.8 trillion-parameter model described in the source as the largest open-weight model to date. The company said full weights are expected to go live before July 27 under a Modified-MIT license, allowing self-hosting. But the hardware bar is far above consumer setups. In Moonshot AI’s deployment guidance, running K3 on your own requires at least 64 accelerators. Even before inference begins, storing the model’s native 4-bit MXFP4 weights takes roughly 1.4TB to 1.5TB. Based on raw weight size alone, that works out to about 19 NVIDIA H100 80GB GPUs, 11 H200s, or 8 B200s, and that does not include KV cache for long-context operation. The article also cites H100 pricing at roughly $25,000 to $33,000 per PCIe card, with SXM versions above $35,000 to $40,000, while an 8-GPU H100 server costs more than $300,000. At Moonshot AI’s stated 64-GPU threshold, hardware procurement alone would exceed $2.4 million. Power use is also heavy: 64 H100 GPUs at about 700W each add up to roughly 45kW, pushing deployment into data-center territory rather than home or hobbyist environments.

Tether unveils QVAC, an open-source, cross-platform local AI framework inspired by Asimov's psychohistory. It treats intelligence as the company's second reserve asset. Early MedPsy models show competitive medical benchmarks at edge scale, challenging centralized cloud AI.

Tether is expanding from stablecoins into AI infrastructure through QVAC, a local-first and peer-to-peer AI stack. Its MedPsy medical models aim to prove that smaller edge-deployable models can outperform larger rivals, but independent replication remains the key test.

Tether is expanding beyond stablecoins into AI infrastructure with QVAC, a local-first, peer-to-peer platform focused on privacy, offline inference, and edge deployment. Its MedPsy medical models are now the first major test of that strategy.

Tether's QVAC project draws on Isaac Asimov's concept of psychohistory from the Foundation series to build a decentralized, local-first AI infrastructure. The article details how Tether leverages its stablecoin reserves (Q1 2026 net profit $1.04B, reserve buffer $8.23B) to fund AI development, the architectural distinction of QVAC: edge-native, peer-to-peer inference, cross-platform SDK. The first model, MedPsy, outperforms Google’s MedGemma-27B on medical benchmarks despite being 7x smaller (1.7B/4B parameters), demonstrating the potential of domain-specific edge-scale models. It also explores the trade-off between convenience and control in local vs. cloud AI, and the critical need for independent replication.