Analysis: Chinese Models Dominate Token Share Yet Pose Limited Revenue Threat to Anthropic
English summary
Chinese open-source large models now account for a majority of token consumption on platforms like OpenRouter, with one week in July seeing 64% of US users' tokens from Chinese models. However, these platforms primarily serve developers and startups, excluding large enterprise contracts and subscriptions, which limits their reflection of overall commercial revenue. Anthropic's annualized revenue growth has slowed, but this is largely due to OpenAI's resurgence rather than Chinese model competition; estimates place Anthropic's ARR at $74.1 billion in mid-July. In high-complexity tasks, Claude still maintains an edge, and per-solved-task pricing narrows the cost gap. The primary contribution of Chinese models has been to accelerate agentic coding through cost-effective offerings, with limited direct revenue impact on Anthropic.
Chinese summary
国产开源大模型在OpenRouter等平台的Token消耗中占据主导，七月某周美国用户64%的Token来自中国模型。但该平台主要服务开发者和初创公司，不包括企业大单和订阅，无法反映整体商业收入。Anthropic的年化收入增速放缓主要归因于OpenAI追赶而非国产模型冲击，7月中旬ARR估值为741亿美元。在复杂任务上Claude仍具优势，按任务解决成本计算价差缩小。国产模型的真正贡献在于以高性价比加速了智能体编程进程，对Anthropic的收入冲击有限。
Key points
Chinese open-source models dominated OpenRouter token consumption (64% of US user tokens in a July week), but this metric excludes enterprise deals and subscriptions.
国产开源模型在OpenRouter平台Token消耗份额极高（七月某周美国用户占比64%），但该数据不含企业合同和订阅。
Anthropic's ARR growth slowed (estimated $74.1B in July, 58% growth in two months), mainly due to OpenAI's resurgence, not Chinese model competition.
Anthropic的ARR增速放缓（7月估值741亿美元，两个月环比增长58%），主要因OpenAI复苏而非国产模型。
In complex tasks, Claude maintains a lead with 65.6% success rate vs. 46.7% for open-source models, and per-solved-task cost narrows the pricing gap.
在复杂任务中Claude成功率65.6%，开源模型46.7%，按任务解决计算成本价差缩小。
Chinese models have accelerated agentic coding through cost-effectiveness, but direct revenue impact on Anthropic remains limited.
国产模型以高性价比加速了智能体编程，但对Anthropic直接收入冲击有限。