A growing wave of developers is publicly announcing cancellations of their subscriptions to frontier AI services, declaring that open-source AI has caught up. While some of this may be performative, the underlying data confirms a real shift in token volume usage.
Tracking signals from expert communities and app usage charts reveals that open-source AI is rapidly gaining traction. For example, DeepSeek’s V4 Flash model now processes over a third of all tokens on Vercel’s infrastructure, surpassing many established providers. In comparison, Anthropic’s flagship Opus 4.8 handles just over 2 trillion tokens weekly, a ratio that was unimaginable a year ago.
New open-source models continue to emerge under permissive licenses, including Tencent’s Hy3 and Z.ai’s GLM 5.2, which leads open-weight intelligence rankings. Other notable models include Moonshot’s Kimi K2.7, Alibaba’s Qwen, and Europe’s Mistral, all offering downloadable weights that rival frontier quality at a fraction of the cost.
Token Volume vs Revenue
Despite the surge in open-source token volume, revenue remains concentrated with frontier labs. Anthropic, for instance, still accounts for more than half of AI spending on Vercel, charging roughly 23 times more per token than DeepSeek. This means that while open-source models dominate production workloads, frontier providers maintain their lead on discovery and the most challenging AI problems.
Decagon’s CEO summarises this dynamic well: frontier labs own discovery, while open source owns production. As workflows mature and stabilise, they migrate to cheaper, "good enough" models, preserving frontier labs' pricing power on the cutting edge.
Implications for AI Pricing and Adoption
Developers testing GLM 5.2 against Anthropic’s Opus found near-indistinguishable performance at under a fifth of the cost. With compatibility across Anthropic and OpenAI endpoints, switching to open-source alternatives can be as simple as a configuration change, accelerating adoption.
This shift exposes frontier labs’ gross margins, which can approach 90% on inference. While this does not threaten their revenue immediately, it signals an opportunity for competitors to erode pricing power over time.
It is important to note that the moat for frontier labs is not the model weights alone but includes distribution, tooling, trust, and the ability to solve the newest, hardest problems. The "goodbye Claude" trend is an early indicator of market evolution rather than an immediate disruption.
As open-source AI continues to mature, businesses should monitor these shifts carefully and consider integrating cost-effective models where appropriate. For expert guidance on AI adoption and automation, explore https://jasonjuul.com, which offers insights grounded in 30+ years of programming expertise.
Disclaimer: This article provides an overview based on current AI market trends and does not constitute financial or investment advice. Readers should conduct their own research before making decisions.