Mozilla Report: Open Model Performance Gap Narrows to ~4 Months, Harness Competition Intensifies
Key point
According to a Mozilla report, the performance gap between open and closed models has narrowed to an average of about 4 months, with competition shifting from the models themselves to harnesses.
Details
Mozilla's 2026 Open Source Status Report suggests that open models are rapidly closing the performance gap with closed models. Based on the Epoch Capabilities Index, the gap between the top open model, Kimi K3 (157 points), and the closed model Claude Opus 5 (162 points), is 5 ECI, which is narrower than the average gap of 8 ECI (approximately 4 months) measured by Epoch. According to the Artificial Analysis Intelligence Index, 4 of the top 10 models are open models (including Kimi K3, GLM-5.3, etc.).
Intensifying Harness Competition
As performance competition among the models themselves eases, competition over harnesses for developer workflows and agent operations is intensifying. Based on Terminal-Bench 2.0, the score difference between third-party scaffolds and proprietary harnesses was 21.8 points, but 8 weeks later in Terminal-Bench 2.1, the gap compressed to approximately 3 points. This shows that labs are internalizing their own harnesses to gain a performance advantage. It was also noted that all published frontier scores depend on the combination of model and harness, meaning simple model comparisons involve many variables.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.