◆ Daily Global-Macro Brief LIVE
⌕ /
TODAY IN 30 SECONDS
  1. Nous Research releases open-source NousCoder-14B coding model, challenging proprietary AI tools
January 7, 2026 1 stories · refreshed 2026-10-12 02:20 UTC ▲ 1

Tech/AI 1

  • NousCoder-14B achieves 67.87% accuracy on LiveCodeBench v6, outperforming base model Qwen3-14B by 7.08pp
  • Model trained in 4 days using 48 Nvidia B200 GPUs, contrasting human sample efficiency (24k vs ~1k problems)
  • Open-sourced full training stack (Atropos framework) enabling replication and extension by researchers
  • Arrives amid Anthropic's Claude Code dominance, highlighting rapid AI coding tool competition
  • LiveCodeBench v6 tests models on competitive programming problems published Aug 2024–May 2025
Read on VentureBeat AI ↗