Mozilla report measures the gap at 4.4 months; Kimi K3 scores three points below Fable 5 at 30% of the cost; Mozilla recommends open models as default for most workloads.
Following·Day 5·first covered 11 Sep 2026·2 sources
Open-Weight Models Now Trail Closed Frontier AI by ~4 Months
The gist
The narrowing gap changes the cost-benefit calculus for organizations choosing between open and closed models. Mozilla's finding that open models cost roughly 30% as much while trailing by only 4.4 months in performance suggests most routine workloads no longer require closed model premiums.
The full picture
A Mozilla report finds the performance gap between closed frontier AI models and the best open-weights models has narrowed to 4.4 months, with leading open models costing roughly 30% as much as closed alternatives. Moonshot AI's Kimi K3 scores three points below Anthropic's Fable 5 on the Artificial Analysis Intelligence Index while costing 30% as much. Mozilla's State of Open Source AI report recommends most organizations use open models as the default for the majority of their work, reserving closed models for expert professional work, high-intensity retrieval, and long-context tasks. Independent analyses from SemiAnalysis corroborate the narrowing gap, with open models progressively closing the distance to the closed frontier. Leading open models have come primarily from Chinese labs since approximately 2024, with some Chinese labs open-sourcing models within hours of release.
How it developed
Independent analyses cited in open-models reading list estimate the open-closed performance gap at 4-6 months, with Chinese labs leading open-weights releases since ~2024.
Sources
Interconnects
Related
- Grew out ofOpenAI's Astra at the Critical cyber tier
- Grew out ofOpenAI's GPT-6 Astra model
The daily email
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free