For a brief, beautiful moment in January 2025, open-weight AI models caught up to their closed-source rivals. The Elo rating gap on Arena’s crowdsourced leaderboard hit zero. Parity.
That moment is over. As of September 2026, the gap between the best closed frontier models and their top open-weight competitors has ballooned to 29 Elo points, according to Arena AI’s evaluation data. Claude Opus 5 Max sits at a rating of 1505, while Moonshot AI’s Kimi K3 Max, the strongest open-weight contender, trails by nearly 30 points. Peter Gostev, Arena’s AI Capability Lead, has been at the center of tracking and visualizing this divergence.
What the numbers actually mean
The trajectory tells a more interesting story than any single snapshot. From zero in January 2025 to 29 points in September 2026, the trend line is moving in the wrong direction for open-weight advocates. Epoch AI’s analysis reinforces this picture from a different angle: open-weight models now lag their closed counterparts by roughly four months in performance on the most challenging tasks. That’s up from a three-month lag observed through late 2025.
Arena processes over 10 million evaluations monthly, drawing on a massive pool of real user interactions rather than synthetic benchmarks. When millions of anonymous users consistently prefer one model’s outputs over another’s, that signal is harder to dismiss.
The business of measuring AI
Arena itself has become a fascinating business story. What started as a UC Berkeley research project in 2023 has transformed into an enterprise pulling in $100 million in annualized revenue. It reached that run-rate within just eight months of launching paid evaluation services.
Gostev’s specific contribution has centered on making model weaknesses legible. His work highlights the tension between how models perform when evaluated by domain experts versus general users, a distinction that matters enormously for enterprise deployments.
The geopolitics of open models
The most active open-weight model development has shifted heavily toward Chinese labs. Moonshot AI’s Kimi series, Zhipu’s GLM, DeepSeek, and Alibaba’s Qwen represent the frontier of what’s publicly available. Yet the gap persists. Anthropic, OpenAI, and Google DeepMind continue to push their closed models further, faster.