Open-weight models at near-frontier quality

Open-weight models have narrowed the quality gap with leading proprietary models to roughly three points on common benchmarks while costing 3x to 6x less. Kimi K3 reaches 95% of Fable 5's intelligence score at 35% of the cost, and GLM-5.2 delivers 85% at 17% of the cost, according to Fastino.ai. V4 Pro scores 80.6% on SWE-bench Verified, placing it in the same range as Claude Opus 4.6 and Gemini 3.1 Pro for coding tasks. OpenRouter reports that open-weight models have maintained a consistent 3-6 month capability gap behind frontier labs for over 18 months, with frontier labs not widening their lead despite earlier expectations. NVIDIA's Nemotron model family achieved gold-level results on both the International Olympiad in Informatics and the International Mathematical Olympiad after fine-tuning. Telecom operators represent a concrete adoption wave: 89% of respondents in NVIDIA's State of AI in Telecommunications report say open-source models are important to their AI strategy, and NVIDIA launched a 30-billion-parameter Nemotron 3 Large Telco Model fine-tuned by AdaptKey on telecom datasets for network configuration and customer incident triage.