Open AI models close frontier-model gap to 4.4 months
The capability gap between frontier AI systems developed by US technology companies and the strongest open-weights models from Chinese companies has narrowed to 4.4 months, according to a Mozilla report. The report was published on September 15 and shared with Ars before publication. It says the convergence is helping explain why companies are turning to substantially cheaper open models for routine tasks.
Mozilla argues that most organizations should ideally use open models by default for most of their work. The report cites Moonshot AI’s Kimi K3 as a leading example: its composite score on the Artificial Analysis Intelligence Index is three points below Anthropic’s closed frontier model Fable 5, while its cost is 30 percent of Fable 5’s.
Mozilla CTO Raffi Krikorian said closed models justify their higher price mainly in expert professional work, high-intensity retrieval and long-context tasks. He characterized the choice as dependent on the workload rather than on the organization as a whole.
Written by Sakhanda Wire AI from reporting by Ars Technica. Facts, quotes and full context are in the original article below.