DeepSeek’s models accounted for 56.4% of the tokens processed by the 10 most-used models on OpenRouter in the week to November 7, up from 52.0% a week earlier, by our arithmetic from the OpenRouter rankings page as our dataset watcher captured it.
The two snapshots were taken at 20:46 UTC on October 31 and 20:12 UTC on November 7. Each shows OpenRouter’s weekly table of the top 10 models, with the tokens each processed.
Business Pill · Tokens Explained
An evergreen Business Engineer explainer on tokens, the unit AI models read, write and bill in. It teaches the general idea only and does not discuss OpenRouter or any company named here.
The key insight: By our arithmetic from OpenRouter’s weekly rankings, DeepSeek’s models made up 56.4% of the tokens processed by OpenRouter’s 10 most-used models in the week to November 7, up from 52.0% a week earlier.
Update, 8 Nov, 19:24 UTC: Kimi K3.1 Pro enters the top 10. In our collection of OpenRouter’s public rankings page, captured at 19:24 UTC on 8 November, the table OpenRouter labels “Top models today” shows Moonshot AI’s Kimi K3.1 Pro at No. 7 with 824B tokens, a change of +56% as shown in OpenRouter’s own change column. Kimi K3.1 Pro was not in that table’s top 10 in our previous capture, at 20:12 UTC on 7 November. In the 8 November capture it sits one place above Anthropic’s Claude Opus 5.5 (No. 8, 767B tokens, −27%).
In the November 7 table, DeepSeek V4.1 Flash 1017 ranks first with 56.2T tokens, up 23% on OpenRouter’s own comparison, and DeepSeek V4.1 Flash ranks second with 17.2T, up 13%. GLM 5.3 Flash from Z.ai is third with 14.1T, down 13%.
OpenAI’s GPT-6 Luna is fourth with 7.22T tokens, down 34%. Tencent’s Hy4 enters at fifth with 7.05T, marked new. Claude Opus 5.5 from Anthropic is seventh with 6.59T, down 10%. StepFun’s Step 5 Preview, MiniMax M3.1 Flash, Tencent’s Hy4 preview and OpenAI’s GPT-5.6 Luna fill the rest of the list.
In the October 31 table, DeepSeek V4.1 Flash 1017 was first with 45.5T tokens, GLM 5.3 Flash second with 16.1T and GPT-6 Luna fourth with 10.9T. Claude Opus 5.5 was sixth with 7.31T.
Share of tokens processed by OpenRouter’s 10 most-used models in the trailing week, by model maker, from OpenRouter’s rankings page as captured on 31 October (20:46 UTC) and 7 November (20:12 UTC). Shares are our arithmetic: each maker’s tokens divided by the sum of the top-10 models’ tokens. Other Chinese labs are Tencent, StepFun and MiniMax.
Our Numbers
We added up the tokens of the 10 models in each table: 124.2T in the week to October 31 and 130.0T in the week to November 7. All shares below are our arithmetic on those totals.
DeepSeek’s three listed models made up 52.0% of the October 31 total; its two listed models made up 56.4% of the November 7 total, a gain of 4.4 points. Z.ai fell from 16.7% to 10.8%, partly because its GLM 5.3 model, ninth on October 31 with 4.62T, is not in the November 7 table.
OpenAI and Anthropic together went from 14.7% to 13.7%. OpenAI held near 8.7% because GPT-5.6 Luna joined the list while GPT-6 Luna fell; Anthropic moved from 5.9% to 5.1%. Tencent, StepFun and MiniMax together rose from 16.6% to 19.0%.
By the same arithmetic, DeepSeek V4.1 Flash 1017 alone processed 56.2T tokens, about 4.1 times GPT-6 Luna and Claude Opus 5.5 combined.
As we read the tables, the gain came from volume: DeepSeek V4.1 Flash 1017 went from 45.5T to 56.2T tokens while GPT-6 Luna went from 10.9T to 7.22T.
As we read them, OpenAI and Anthropic together slipped from 14.7% to 13.7% of the top 10, and Tencent, StepFun and MiniMax together rose from 16.6% to 19.0%.
As we read the scope, these are one platform’s top-10 tables in rolling weeks; a model entering or leaving the list moves the shares.
Tokenomics: The Economics of AI, The Business Engineer
“It is simultaneously the unit of cognition the model produces, the unit of compute the data center serves, the unit of price the lab charges, and the unit of value the enterprise extracts.”
Three Implications
FOR DEEPSEEK OpenRouter’s November 7 table lists its two V4.1 Flash models first and second, with 56.2T and 17.2T tokens.
FOR OPENAI AND ANTHROPIC OpenRouter’s table shows GPT-6 Luna down 34% and Claude Opus 5.5 down 10% on its own weekly comparison.
FOR READERS OF THESE TABLES The shares cover OpenRouter’s top 10 only and exclude direct API traffic to any lab.
The analysis says of the token: “It is simultaneously the unit of cognition the model produces, the unit of compute the data center serves, the unit of price the lab charges, and the unit of value the enterprise extracts.”
As we read it, OpenRouter’s weekly table counts the first of those units, tokens processed, by model maker. The analysis does not discuss OpenRouter or these rankings.
What Is Not Established
These are OpenRouter’s tables, not the whole market: direct API traffic to OpenAI, Anthropic, Google or DeepSeek is not included. The shares cover only the top 10 models in each week, so a model leaving or entering the list moves the figures. The tables are rolling seven-day windows captured at a moment. We make no claim about revenue.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
On OpenRouter’s weekly table, DeepSeek’s models rose to 56.4% of the top-10 models’ tokens in the week to November 7, from 52.0% a week before, by our arithmetic. GPT-6 Luna’s volume fell 34% and Claude Opus 5.5’s fell 10% on OpenRouter’s own figures. The data covers one platform and its top 10 only.
A note on sourcing. Model names, ranks, token counts and percentage changes are from OpenRouter’s rankings page, weekly top-10 tables captured by our dataset watcher on October 31 and November 7, 2026. Totals and shares are our arithmetic. This piece is not investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
