OpenRouter, the AI modelmarketplace Stripe agreed to buy (our piece), says it routed 790 trillion tokens in October, up 37% from September and about 25x from January. Agents consumed 76% of them, about 600 trillion, against 8% for agentic API keys a year earlier (OpenRouter).
The figures cover traffic through OpenRouter only, not the whole AI market. OpenRouter published the recap on 6 November, with a thread from its own account (OpenRouter on X).
Business Pill · AI Agents
An evergreen Business Engineer explainer on AI agents. It teaches the general idea only and does not discuss OpenRouter.
The key insight: OpenRouter says it routed 790 trillion tokens in October, about 25x January, with agents using 76% of them and effective prices per token down 57% to 70% since January; the data covers OpenRouter’s own traffic.
OpenRouter calls October “the largest month in OpenRouter’s history”. It says the refresh of DeepSeek V4.1 Flash and the arrival of Claude Haiku 5.5 and the rest of the Claude 5.5 family “were major contributors”.
It classifies every API key as agentic, human or mixed with a classifier of seven signals. Because agents resend long context on every turn, it says the median provider cache hit rate on agentic traffic rose from 76.1% in June to over 90% in October.
The effective price per million tokens fell from $1.21 to $0.52 for closed-weight models and from $0.28 to $0.084 for open weights since January, and the gap between them widened from 4.3x to 6.2x, OpenRouter says. It notes the effective price can move with list prices, the model mix or cache hits, not only price cuts.
OpenRouter’s blended price per million tokens, January vs October 2026: closed-weight models fell from $1.21 to $0.52 and open-weight models from $0.28 to $0.084. Data from OpenRouter’s October recap, 6 November 2026, covering OpenRouter traffic only; percent changes are our arithmetic.
What Was Posted on X
Our October data recap covers token growth, agentic usage, token prices, model lab retention, and inference providers. Here’s what stood out.
1/ October in tokens: 790T tokens, up 25x from January.
OpenRouter’s own thread at 20:27 UTC on 6 November, opening its October data recap: “790T tokens, up 25x from January.” Later posts in the thread put agents at 76% of all tokens and give the price, retention and provider figures this piece uses.
Monthly AI report from OpenRouter.
Major stuff:
– Agents now use 75% of tokens
– Price per million tokens down at least 57% from Jan
– DeepSeek and Zai lead the charts in both token growth and spend growth since May
Peter Walker, Head of Insights at OpenRouter per the X bio, summarising the recap at 21:21 UTC: agents at “75% of tokens” and prices “down at least 57% from Jan”. OpenRouter’s thread gives the agentic share as 76%.
Open Weights, Labs and Retention
OpenRouter says the open-weight share of paid tokens fell to 52% in the week of 28 September, when Claude Opus 5.5, Sonnet 5.5 and GPT-6.1 Sol launched, and was back above 70% two weeks later, “mostly because of DeepSeek V4.1 Flash”. Open weights take about a quarter of paid spend, up from 11% in January, it says.
Since May, it says Z.ai’s monthly tokens grew about twentyfold and DeepSeek’s about twelvefold; on spend, Z.ai grew about sixfold and Anthropic rose 76%. Of accounts with 100+ paid requests to a lab in September, it says Google kept 78% in October, and DeepSeek, Anthropic and OpenAI 74% each.
By tokens, DeepSeek V4.1 Flash leads most coding and agent tasks on its task leaderboard; by spend, Claude Opus 5.5 leads most of the board. Decision models reached 21% of all requests by the week of 26 October, a category that started in mid-September. Among open-weight inference providers, Together rose from 18th in June to first by tokens served, OpenRouter says.
Our Numbers
790 trillion tokens over October’s 31 days is about 25.5 trillion a day, by our arithmetic. A 25x rise over the nine months from January implies about 43% compound growth a month, and 37% growth into October implies about 577 trillion tokens in September, also by our arithmetic. At October’s pace, the year runs at about 9.5 quadrillion tokens, close to the run rate in our 2 November piece.
Connecting the Coverage
We reported in October that agents burn nearly 5x the tokens people do on OpenRouter (our piece) and that Anthropic tokens on OpenRouter rose 3.6x after Opus 5.5 (our piece). DeepSeek’s own agent reached 1.54M daily users (our piece), Nous retuned Alibaba’s Qwen (our piece), and OpenAI priced GPT-6 Cyber at 2x Astra (our piece).
The Structural Read
As we read the recap, agents now set the volume: OpenRouter says agentic keys took 76% of tokens in October, against 8% a year earlier.
As we read the prices, the gap is widening: OpenRouter says closed-weight tokens cost 4.3x open-weight tokens in January and 6.2x in October.
As we read the retention table, the big labs keep users at similar rates: OpenRouter says Google kept 78% of September accounts and DeepSeek, Anthropic and OpenAI 74% each.
OpenRouter, 6 November 2026
“the largest month in OpenRouter’s history”
Three Implications
FOR LABS OpenRouter says Z.ai’s monthly tokens grew about twentyfold and DeepSeek’s about twelvefold since May, while Anthropic’s spend rose 76%.
FOR INFERENCE PROVIDERS OpenRouter says the median provider cache hit rate on agentic traffic rose from 76.1% in June to over 90% in October.
FOR OPEN WEIGHTS OpenRouter says open weights take about a quarter of paid spend, up from 11% in January.
The analysis says: “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 October recap measures the token as volume and as price at once; the analysis does not discuss OpenRouter.
What Is Not Established
All figures are OpenRouter’s own and describe one router’s customers, who skew toward developers and open-weight models. They do not show lab-wide revenue or usage, and OpenRouter’s agentic classification is its own method. Effective prices reflect the mix of models used, not only list prices.
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.
OpenRouter says it routed 790 trillion tokens in October, 25x January, with agents using 76% of them and effective prices down 57% to 70% since January. The data covers OpenRouter’s own traffic.
A note on sourcing. All figures are OpenRouter’s own, from its October 2026 data recap and its X thread of 6 November, and cover OpenRouter’s traffic only. Daily, monthly-growth, September and annual figures are our arithmetic. This 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.
