State of AI: An Empirical 100T Token Study with OpenRouter
openrouter.ai184 points by anjneymidha 12 hours ago
184 points by anjneymidha 12 hours ago
Super interesting data.
I do question this finding:
> the small model category as a whole is seeing its share of usage decline.
It's important to remember that this data is from OpenRouter... a API service. Small models are exactly those that can be self-hosted.
It could be the case that total small model usage has actually grown, but people are self-hosting rather than using an API. OpenRouter would not be in a position to determine this.
Thank you & totally agree! The findings are purely observational through OpenRouter’s lens, so they naturally reflect usage on the platform, not the entire ecosystem.
While it is possible to self-host small models, it is not easy to host them with high speeds. Many small-model use-cases are for large batches of work (processing large amounts of documents, agentic workflows, ...), and then using a provider that has high tps numbers would be motivated.
Still, I agree that self-hosting is probably a part of the decrease.
The bigger issue is that they count small based on fixed number of parameters, and not the active parameter for MoE, didn't account for any hardware improvements etc. If they counted small based on the price or computational cost, I think they would have seen increase in small models.
I like to see stats like that, but I find it very concerning that OpenRouter don't mind inspecting its user/customer data without shame.
Even if you pretend that the classifier respect anonymity, if I pay for the inference, I would expect that it would be a closed tube with my privacy respected. If at least it was for "safety" checks, I don't like that but I would almost understand, now it is for them to have "marketing data".
Imagine, and regarding the state of the world it might come soon, that you have whatsapp or telegram that inspect all the messages that you send to give reports like:
- 20% of our users speak about their health issues
- 30% of messages are about annoying coworkers
- 15% are messages comparing dick sizes
They explicitly give you a discount if you opt in to allowing your data to be used for (anonymized) analytics. That’s pretty fair imho.
Cynical take: they could look at everyone and give a discount for optics.
I'd feel a lot better if "OpenRouter" were open source.
litellm is basically open source version https://www.litellm.ai although the openrouter being a hosted service is kinda the point. Unless the whole industry decides to this over e2ee you cant get any guarantees about an intermediary aggregator
> inspecting its user/customer data without shame
Even better: they send all the data to GoogleTagClassifier, which means now Google had a copy of the sample as well
>I would expect that it would be a closed tube with my privacy respected
Lol hahaha
Comments such as these are what allows developers to justify ignoring our privacy.
"Everyone knows in {{BUSINESS_TYPE}} there is no real privacy".
Be it fintech, AI or social media. You give them a free pass with being flippant about companies respecting privacy.
Being flippant about anyone being careless about our privacy is doing us as a society and injustice. We should demand privacy, not laugh at the notion of privacy.
Very interesting how Singapore ranks 2nd in terms of token volume. I wonder if this is potentially Chinese usage via VPN, or if Singaporean consumers and firms are dominating in AI adoption.
Also interesting how the 'roleplaying' category is so dominant, makes me wonder if Google's classifier sees a system prompt with "Act as a X" and classifies that as roleplay vs the specific industry the roleplay was intended to serve.
Almost certainly VPN traffic. Most major LLMs block both China and Hong Kong (surprisingly, not the other way around), so Singapore ends up being the fastest nearby endpoint that isn't restricted.
Why do major LLMs block china? Isn't that a potentially huge market for them?
Early on there was a lot of distillation going on, apparently. Note that OpenAI introduced ID verification for high volume accounts and I think it was for that reason. It does raise questions about how much of the Chinese model's performance is entirely home grown. At least historically, it was quite hard to crawl the English web from behind the Great Firewall.
I'm not sure, but my guess is that it's due to pressure (or perceived pressure) from the U.S. government.
It’s not VPN traffic all data is aggregated by billing payment information so it’s Singaporean billing details.
Ah, you're right. Still, I wonder if it's because of Chinese people and companies using Singaporean bank accounts. It just seems odd that such a small country is so overrepresented here.
According to the report, 52% of all open-source AI is used for *roleplaying*. They attribute it to fewer content filters and higher creativity.
I'm pretty surprised by that, but I guess that also selects for people who would use openrouter
Or maybe it’s just strange classification. I see a lot of prompts on the internet looking like “act as a senior xxx expert with over 15 years of industry experience and answer the following: [insert simple question]”
I hope those are not classified as “roleplaying” the “roleplay” here is just a trick to get better answer from the model, often in a professional setting that has nothing to do with creative writing of NSFW stuff
I can't be sure, but this sounds entirely possible to me.
There are many, many people, and websites, dedicated to roleplaying, and those people will often have conversations lasting thousands of messages with different characters. I know a people whose personal 'roleplay AI' budget is a $1,000/month, as they want the best quality AIs.
Would be good to look into those particular statistics, then. Seems like the category could include all sorts of stuff:
> This indicates that users turn to open models primarily for creative interactive dialogues (such as storytelling, character roleplay, and gaming scenarios) and for coding-related tasks. The dominance of roleplay (hovering at more than 50% of all OSS tokens) underscores a use case where open models have an edge: they can be utilized for creativity and are often less constrained by content filters, making them attractive for fantasy or entertainment applications. Roleplay tasks require flexible responses, context retention, and emotional nuance - attributes that open models can deliver effectively without being heavily restricted by commercial safety or moderation layers. This makes them particularly appealing for communities experimenting with character-driven experiences, fan fiction, interactive games, and simulation environments.
I could imagine something like D&D or other types of narrative adventures on demand with a machine that never tires of exploring subplots or rewriting sections to be a bit different is a pretty cool thing to have. Either that, or writing fiction, albeit hopefully not entire slop books that are sold, but something to draw inspiration from and do a back and forth.
In regards to NSFW stuff, a while back people were clowning on OpenAI for suggesting that they'd provide adult writing content to adults, but it might as well be a bunch of money that's otherwise left on the table. Note: I'm all for personal freedom, though one also has to wonder about the longer term impact of those "AI girlfriend/boyfriend" trends, you sometimes see people making videos about those subreddits. Oh well, not my place to judge.
Edit: oh hey, there is more data there after all
> Among the highest-volume categories, roleplay stands out for its consistency and specialization. Nearly 60% of roleplay tokens fall under Games/Roleplaying Games, suggesting that users treat LLMs less as casual chatbots and more as structured roleplaying or character engines. This is further reinforced by the presence of Writers Resources (15.6%) and Adult content (15.4%), pointing to a blend of interactive fiction, scenario generation, and personal fantasy. Contrary to assumptions that roleplay is mostly informal dialogue, the data show a well-defined and replicable genre-based use case.
OpenRouter classifies content by the app that's used to interact with the llm. https://openrouter.ai/docs/app-attribution
"act as a senior xxx expert with over 15 years of industry experience"
... I just don't get why LLMs are affected by this kind of nonsense -- is it due to training rewards?
The way I think about it, the training data (i.e. the internet) has X% of people asking something like "explain it to me like I'm five years old" and Y% of people framing it like "I'm technical, explain this to me in detail". You use the "act as a senior XXX" when you want to bias the output towards something more detailed.
Openrouter has an apps tab. If you look at the free, non-coding models, some apps that feature are: janitor.ai, sillytavern, chub.ai. I'd never heard of them but people seem to be burning millions of tokens enjoying them.
If you rely on AI to write most of your code (instead of using it like Stackoverflow), Claude Code/OpenAI Codex subscription are cheaper than buying tokens. So those users are not on openrouter.
I'm curious what percentage of claude/codex users this is true for - I assumed their business models rely on this not being true for the majority.
Both Claude Code and Codex steer you towards the monthly subscription. Last time I tried Codex, I remember several aspects of it being straight up broken if used with an API key instead of a subscription account.
The business model is likely built upon the assumption that most people aren't going to max out their limits every day, because if they were, it likely wouldn't be profitable.
I got $250 free Claude Code credit and I was surprised by how hard it was to actually use it all before it expired.
That also stuck out for me, I was wondering if it was video games using openrouter for uptime / inference switching, video games would use a lot of tokens generating dialogue for a few programmer's villages.
> I guess that also selects for people who would use openrouter
It definitely does. OpenRouter is pretty popular among roleplayers and creative writers due to having a wide variety of models available, sometimes providing free access to quality models such as DeepSeek, and lacking any sort of rules against generating "adult" content.
> The noticeable spike [~20 percentage points] in May in the figure above [tool invocations] was largely attributable to one sizable account whose activity briefly lifted overall volumes.
The fact that one account can have such a noticeable effect on token usage is kind of insane. And also raises the question of how much token usage is coming from just one or five or ten sizeable accounts.
It is quite interesting to ponder these usage statistics, isn't it?
According to their charts they're at a throughput of something like 7T tok/week total now. At 1$/Mtok, that's 7M$ per week. Less than half a billion per year. How much is that compared to the total inference market? And yet again, their throughput went like 20x in one year, who knows what's to come...
Yes, but that token growth chart looks linear to me. There's the usual summer slump and then growth catches up once the autumn begins, but if you plot a line from the winter growth period at the start of 2025 you end up roughly in the right place except for an unusual spike in the most recent month (maybe another big user).
I'd have liked to see a chart of all tokens broken down by category rather than just percentages, but what this data seems to be saying is that growth isn't exponential, and is being dominated by growth in programming. A lot of the spending in AI is being driven by the assumption that it'll be used for everything everywhere. Perhaps it's just OpenRouter's user base, but if this data is representative then it implies AI adoption isn't growing all that fast outside of the tech industry (especially as "science" is nearly all AI related discussion).
This feels intuitively likely. I haven't seen many obvious signs of AI adoption around me once I leave the office. Microsoft has been struggling to sell its Copilot offerings to ordinary MS Office users, who apparently aren't that keen. The big wins are going to be existing apps and data pipelines calling out to AI, and it'll just take time to figure out what those use cases are and integrate them. Integrating even present-day AI into the long tail of non-tech industries is probably going to take decades.
Also odd: no category for students cheating on homework? I notice that "editing services" is a big chunk of the "academia" category. Probably most of that traffic goes direct to chatgpt.com and bypasses OpenRouter entirely.
These are fantastic insights! I work in legaltech space so something to keep in mind is that legal space is very sensitive to data storage and security (apart from this of course: https://alexschapiro.com/security/vulnerability/2025/12/02/f...). So models hosted in e.g. Azure, or on-prem deployments are more common. I have friends in health space and similar story there. Finance (banking especially) is the same. Hence why those categories look more or less constant over time, and have smallest contributions in this study.
I worry that OpenRouter's Apps leaderboard incentivizes tools (e.g. Cline/Kilo) to burn through tokens to climb the ranks, meanwhile penalizing being context-efficient.
The 'Glass slipper' idea makes sense to me; people have a bunch of different ideas to try on AIs, and try it as new models come out, and once a model does it well they stick with it for a while.
With studies like these it's important to keep in mind selection effects.
Most of the high volume enterprise use cases use their cloud providers (e.g., azure)
What we have here is mostly from smaller players. Good data but obviously a subset of the inference universe.
Here is the thing: they made good enough open weight models available and affordable, then found that people used them more than before. I am not trying to diminish the value here but I don’t think this is the headline.
The open weight model data is very interesting. I missed the release of Minimax M2. The benchmarks seem insanely impressive for its size. I would suspect benchmaxing but why would people be using it if it wasn’t useful?
I used it for a couple of days and was very impressed. Definitely a hidden gem, and super cheap.
> The metric reflects the proportion of all tokens served by reasoning models, not the share of "reasoning tokens" within model outputs.
I'd be interested in a clarification on the reasoning vs non-reasoning metric.
Does this mean the reasoning total is (input + reasoning + output) tokens? Or is it just (input + output).
Obviously the reasoning tokens would add a ton to the overall count. So it would be interesting to see it on an apples to apples comparison with non reasoning models.
As would models that that are overly verbose. My experience is the Claude tends to do more than is asked for (e.g. immediately move on to creating tests and documentation) while other models like Gemini tend to be more concise in what they do.
I'm out of time but "reasoning input tokens" from fortune 5000 engineers sounds like a lobotomized LSD dream, would you care on elaborating how you distinguish between reasoning and non-reasoning? vs "question on duty"?
"reasoning" models like GPT 5 et al do a pre-generation step where they:
- Take in the user query (input tokens)
- Break that into a game plan. Ex: "Based on user query: {query} generate a plan of action." (reasoning tokens)
- Answer (output tokens)
Because the reasoning step runs in a loop until it's run through it's action plan, it frequently uses way more tokens than the input/output step.
that was useful, thank you.
I have sooo many issues with the naming scheme of this """""AI"""" industry", it's crazy!
So the LLM gets a prompt, then creates a scheme to pull pre-weighted tokens post-user-phrasing, the constituents of which (the scheme) are called reasoning tokens, which it only explicitly distinguishes as such because there are hundreds or even thousands of output tokens to the hundreds and/or thousands of potential reasoning input tokens that were (almost) equal to the actually chosen reasoning input tokens based on the more or less adequately phrased question/prompt given ... as input ... by the user ...
You can call them planning if you want or pre-planning. But I would encourage you to play with the API version of your model of choice to see exactly what this looks like. It’s kind of like a human’s internal monologue: “got an email from my boss asking to write unit tests for the analytics API. First I have to look at the implementation to know how exactly it actually functions, then write out what kinds of tests make sense, then implement the tests. I should write a TODO list of these steps.”
It is essentially a way to expand the prompt further. You can achieve the same exact thing by turning off the “thinking” feature and just being more detailed and step by step in your prompt but this is faster.
My guess is that the next evolution of this will be models that do an edit or review step after to catch if any of the constraints were broken. But best I can tell a reasoning model can be approximated by doing two passes of a non-reasoning model: first pass you give it the user prompt with instructions that boil down to “make sense of this prompt and formulate a plan” and the second pass you give it the original prompt, the plan, and an explanation that the plan is to implement the original prompt using the plan.
I believe they’re just classifying all models into “reasoning models” eg o3 vs “non reasoning models” eg 4o and just doing a comparison of total tokens (input tokens + hidden reasoning output tokens + shown output tokens)
that's exactly right!
hell yeah, 109 out of 10 doors opened! 99 bonus doors! what are you talking about, man?
Who is using grok code and why?
It was (is?) free with eg. opencode -- so, open-source coding agent + free sota model, it's hard to resist. That said, grok fast is fast, but not that great when compared to the other top tier models.
It's a 1.7 trillion token free model. Why wouldn't you try it?
I've been testing free models for coding hobby projects after I burnt through way too many expensive tokens on Replit and Claude. Grok wasn't great, kept getting into loops for me. I had better results using KAT coder on opencode (also free).
> Why wouldn't you try it?
Because the people behind it and myself having at least some standards
According to https://openrouter.ai/rankings, lots of people are using it - presumably because it performs well and provides value.
Kilo Code lets people use Grok Code Fast 1 for free, using OpenRouter as the provider. And Grok 4.1 Fast was completely free directly on OpenRouter for some time after its release.
So yeah, their statistics are inflated quite a bit, since most of that usage was not paid for, or at least not by the end user.
This is interesting, but I found it moderately disturbing that they spend a LOT of effort up front talking about how they don’t have any access to the prompts or responses. And then they reveal that they did actually have access to the text and they spend 80% of the rest of the paper analyzing the content.
>And then they reveal that they did actually have access to the text
I'm not seeing that. All I'm seeing is them analyzing metadata.
>All I'm seeing is them analyzing metadata Read the section about how they achieve classifications for prompts (hint: They read the prompts)
I didn’t read the paper but I know OR has an option to opt-in to reading/training off the prompts for a discount. Some free models also log, but I’m not sure if that is just the provider, or OR too
I am a person who wants to maintain a distance from the AI-hype train, but seeing a chart like this [1], I can't help think that we are nowhere near the peak. The weekly token consumption keeps on rising, and it's already in trillions, and this ignores a lot of consumption happening directly through APIs.
Nvidia could keep delivering record-breaking numbers, and we may well see multiple companies hit six, seven, or even eight trillion dollars in market cap within a couple of years. While I am skeptical of claims like AI will make programming obsolete, but it’s clear that the adoption is still going like crazy and it's hard to anticipate when the plateau happens.
[1]: https://openrouter.ai/state-of-ai#open-vs_-closed-source-mod...
Problem is that most of that growth is in models being underpriced. We don't know what the demand curve looks like when tokens are priced to cover the full operating costs of the companies making them.
Also, growth seems to be linear, not exponential.
When it's as cheap as 5 cents per million tokens I don't see "trillions" as being a particularly large number. Even at the most expensive level($120/1M for 5 Pro) 100 trillion tokens is only like $12 billion dollars.
Overall really interesting read, but I'm having trouble processing this:
> OpenRouter performs internal categorization on a random sample comprising approximately 0.25% of all prompts
How can you arrive at any conclusion with such a small random sample size?
Statistical significance comes mostly from N (number of samples) and the variance on the dimension you're trying to measure[1]. If the variance is high, you'll need higher N. If the variance is low, you'll need a lower N. The percentage of the population is not relevant (N = 1000 might be significant and it doesn't matter if it's 1% or 30% of the population)
[^1] This is a simplification. I should say that it depends on the standard error of your statistic, i.e, the thing you're trying to measure (If you're estimating the max of a population, that's going to require more samples than if you're estimating the mean). This standard error, in turn, will depend on the standard deviation of the dimension you're measuring. For example, if you're estimating the mean height, the relevant quantity is the standard deviation of height in the population.
https://en.wikipedia.org/wiki/Central_limit_theorem
For example, even 300 really random people is enough to correctly assertain the distribution of population for some measurement (say, some personality feauture).
That’s the basis of all polls and what have you
I think you might be thrashing around 30 samples for a normal distribution and the Central Limit Theorem and accidentally added a zero!
(OK, on rereading, you did link to a WP article about CLT, so 30 it is!)
You’re absolutely right! (c)
300 — I had in memory as a safe bet in a case of some skewed stuff like log-normal, exponential, etc.
Because the accuracy of an estimated quantity mostly depends on the size of the sample, not on the size of the population [1]. This does require assumptions like somewhat homogenous population and normal distributions etc. However, these assumptions often hold.
[1] https://stats.stackexchange.com/questions/166/how-do-you-dec...
Very cool study
This is really amazing data. Super interesting read
*State of non-enterprise, indie AI
All this data confirms that OpenRouter’s enterprise ambitions will fail. It’s a nice product for running Chinese models tho
They have SOTA models from OpenAI and Anthropic and Google and you can access them at a 5.5% premium. What you get is the ability to seamlessly switch between them. And also when one is down you can instantly switch to another. Whether that is valuable to you or not is use case dependent. But it isn’t without value.
What it does have I think is a problem that TaskRabbit had: you can hire a house cleaner through TR but once you find a good one you can just work directly with them and save the middleman fee. So OR is great for experimenting with a ton of models to see what is the cheapest one that still performs the tasks you need but then you no longer need OR unless it is for reliability.