AnalysisAI ModelsJuly 24, 2026
DeepSWE: Kimi K3 delivers 2.8x solves per dollar vs Claude Fable 5
Together AI ran 452 DeepSWE rollouts on both models. Claude Fable 5 leads pass@1 by 1.4 points, but Kimi K3 wins pass@4 and delivers 2.8x the solves per dollar.
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- Daniel Kokotajlo and AI Futures Project release AI 2040: Plan A
- LearnHouse open-source LMS features AI integration and whiteboards
- Practical guide to Claude Code custom slash commands
- Reddit debate over xAI's Grok 4.5 vs Meta's AI spending
- Aravind Srinivas: two orders of magnitude improvements 'a big deal'