Bittensor Subnet Runs Full Kimi K3 on 80 RTX 5090s, Halving API Costs
markjeffrey · x · 2026-08-13
Bittensor's Subnet 53 has launched the full 2.8-trillion-parameter Kimi K3 model, running it on a decentralized network of 80 consumer-grade RTX 5090 GPUs.
The setup delivers performance comparable to Moonshot's official API, while allowing OpenRouter users to save up to 50% in costs. This highlights the potential of decentralized compute networks to significantly reduce LLM inference expenses.
More from Infra
- SK Hynix to Invest $38.1B in Two New Memory Fabs in Korea — Beth_Kindig · 2026-08-13
- Emerging Inference Engine 'Tokenspeed' Gains Day-0 Support for Qwen3.8 — This_Maintenance_834 · 2026-08-13
- DeepSeek V4 Analysis: Ultra-low API Costs and SSD KV Cache Revolution — Xianbao_QIAN · 2026-08-13
- Vercel AI Gateway Adds Grok and DeepSeek Models with Zero Markup — brandon_galang · 2026-08-13
- Rethinking AI Energy Metrics: Joules per Token Must Account for Quality and Task — prateekj · 2026-08-13
- Benchmark: 124B Model Hits 38.7 tok/s on a Single DGX Spark — AcanthisittaOk1699 · 2026-08-13