Open-source AI nearly closes gap with top proprietary models

The distance between the best AI models anyone can download and the industry's top proprietary systems has shrunk to its narrowest margin yet, reshaping how Corporate America thinks about artificial intelligence spending.

According to the Artificial Analysis Intelligence Index for September 2026, the leading open-weight model — Kimi K3 from Moonshot AI — scores 59.7, just 6 points behind the overall leader, Anthropic's Claude Fable 5.1 at 65.7. A year ago, that gap stood at 13 points. Four open models have now crossed the 50-point threshold on the index, a level that none reached twelve months prior.pasqualepillitteri

Corporate America Shifts Away From Costly Closed Models

The benchmark convergence is translating directly into enterprise behavior. AT&T ↗ 0.43%, which processes roughly 45 billion AI tokens daily, has been aggressively expanding its use of open-weight models. By May, open models accounted for 20% of AT&T's AI use; that figure has since risen to 40% and may reach 60% in the coming months, according to Andy Markus, the company's chief data and AI officer. Markus said AT&T is saving up to 80% on AI costs compared with earlier this year.indianexpress+1

The trend extends well beyond one telecom. Open models accounted for 58% of AI use last month, up from 10% a year ago, according to U.S. user data from OpenRouter. Companies including Airbnb ↘ 0.87% and Deloitte are also adopting open alternatives.indianexpress

Nvidia's $12.9 Billion Bet

On Thursday, Nvidia ↗ 0.92% announced it would acquire Hugging Face, the leading library of open AI models, for $12.9 billion. "Together, we will make AI more open, more capable and more accessible to people and institutions around the world," CEO Jensen Huang said. Reuters described the deal as Nvidia using "its growing AI war chest to expand its influence in the booming market for open-source models".reuters+2

A Chinese-Led Open-Weight Tier

Many of the top open models are built by Chinese labs. Moonshot's Kimi K3, a 2.8-trillion-parameter mixture-of-experts model, leads the open field, followed by Z.AI's GLM-5.3 at 59.5, Alibaba's ↗ 0.02% Qwen3.8 Max Preview at 58.1, and DeepSeek V4 Pro at 53.2. Some U.S. firms remain cautious about Chinese models over data-privacy concerns; AT&T, for instance, researches them but relies instead on Google's Alphabet Inc. ↗ 0.85% Gemma and Meta's ↘ 1.64% Llama.pasqualepillitteri+1

As Jerry Tang, CEO of AI startup Atlas Cloud, put it: "Mazdas do a fine job getting you where you need to go, and Maseratis are simply not worth the investment".indianexpress