AnalysisAI ModelsJuly 11, 2026
Hy3 (295B MoE) and Nemotron-Labs-Audex-30B-A3B GGUF quants shared

A Reddit user shared GGUF quantized versions of Hy3 (295B MoE) and NVIDIA's Nemotron-Labs-Audex-30B-A3B (30B audio-capable MoE). The quants use imatrix quantization, with KLD/PPL measurements against BF16 reference logits and llama-bench throughput numbers. All raw benchmark data is included in the repos for reproducibility.
1 source
More stories today
- Seeing AI Agents Is Not Enough. Security Teams Must Enforce What They Can Do
- Proposed 'Genie Coefficient' measures AI alignment gap
- Stripe in talks to acquire OpenRouter for $10 billion
- BrainAPI converts raw text to knowledge graph via agent swarm
- Israel, UK name AI ministers to compete with US, China