The State of Open Source AI report from Mozilla, released on September 15, indicates that open-weights models are increasingly competitive with closed frontier systems. Moonshot AI’s Kimi K3 model trails Anthropic’s Fable 5 by only three points on the Artificial Intelligence Index while costing 30 percent of the price.
Performance gaps are narrowing, with closed models currently holding a four-month lead in handling tasks that take between eight and 12 hours. While closed models remain dominant for professional work and long-context tasks, open models are becoming the default for routine operations due to their cost-effectiveness.
Benchmarking by Vals AI using the Terminal-Bench 2.1 evaluation shows that the GLM 5.2 model from Zhipu AI performs nearly on par with Anthropic’s Claude Opus 4.7 and 4.8, despite costing five times less per task. Data from OpenRouter confirms this trend, with eight of the top 10 models by token volume in August 2026 being open-weights.
Despite the growth in usage, open models generated only 4 percent of revenue compared to 96 percent for closed models between May and September 2025. Mozilla CTO Raffi Krikorian noted that the open-weights ecosystem is currently heavily reliant on Chinese labs and capital, prompting calls for Western institutions to build a more diverse, neutral coalition for AI development.
