Open models have narrowed the gap with proprietary ones to 4 months

The research project State of Open Source AI published a report according to which the performance gap between closed frontier models and open analogs has narrowed to 4 months. Analysis shows that using paid APIs costs companies 5 times more when receiving only a slight advantage in access speed to advanced capabilities. During the research, experts compared metrics of leading proprietary systems with open solutions presented on the Artificial Analysis platform. It turned out that open models, including developments from China, demonstrate comparable results in coding and logical inference tasks. At the same time, the cost of operating open models with self-hosting is significantly lower than the costs of subscribing to cloud frontier solutions. Data from Ars Technica confirms that the pace of fine-tuning and optimization of open architectures allows them to catch up with flagship models from Silicon Valley within one quarter. The main driver of this process was the availability of high-quality datasets and knowledge distillation methods, which were previously the exclusive advantage of large corporations. For business, this means a change in the strategy of choosing infrastructure. If previously the use of closed APIs was the only way to get access to SOTA results, now companies can switch to local or cloud open models without losing quality, saving up to 80% of the budget on inference. The market is moving from a monopoly of proprietary systems to a model where a 4-month advantage becomes the only argument in favor of paid subscriptions.