OpenRouter's rankings update daily with real token volumes from millions of developers. No surveys. No estimates. Just what people are actually using.
The June 2026 data is surprising in ways that matter for investors.
Most AI market analysis focuses on model quality benchmarks, funding announcements, or executive interviews. OpenRouter offers something rarer: a real-time, public scoreboard of which AI models developers are actually running, measured in tokens processed per day.
OpenRouter is a model routing platform that lets developers access any of 400+ AI models through a single API. Its rankings page updates daily. The June 2026 data — broken down by individual model, by company, by use case, and by language — tells a story that is more nuanced than the "China is winning" narrative that has dominated AI headlines.
Here is what the data actually shows.
The Company Scoreboard: Who Is Really Winning
The most important view isn't the individual model rankings — it's the weekly token volume by company, which smooths out single-model spikes and shows structural market position.
Weekly Token Volume by Company (June 2026)
| Rank | Company | Origin | Weekly Tokens | Market Share |
|---|---|---|---|---|
| 1 | DeepSeek | 🇨🇳 China | 5.13T | 17.6% |
| 2 | Anthropic | 🇺🇸 US | 4.34T | 14.8% |
| 3 | 🇺🇸 US | 3.66T | 12.5% | |
| 4 | OpenAI | 🇺🇸 US | 2.46T | 8.4% |
| 5 | Xiaomi | 🇨🇳 China | 2.42T | 8.3% |
| 6 | MiniMax | 🇨🇳 China | 2.37T | 8.1% |
| 7 | Tencent | 🇨🇳 China | 2.36T | 8.1% |
| 8 | OpenRouter | — | 1.53T | 5.2% |
| 9 | Qwen (Alibaba) | 🇨🇳 China | 1.26T | 4.3% |
| 10 | Others | — | 3.7T | 12.7% |
Source: OpenRouter.ai/rankings, Market Share view, June 2026. Total weekly volume: ~29.2 trillion tokens.
Chinese-origin companies (named): DeepSeek + Xiaomi + MiniMax + Tencent + Qwen = 46.4% of all identified token volume.
US-origin companies (named): Anthropic + Google + OpenAI = 35.7%.
Two things stand out immediately. First, DeepSeek — not Xiaomi, not Tencent — is the dominant force with nearly one in five tokens on the platform. Second, Anthropic holds the #2 position at 14.8% despite the narrative that premium US models are being displaced. Those two facts together define the real AI market dynamic in mid-2026.
Today's Model Leaderboard
The daily individual model rankings show the current snapshot of developer choice:
Top 10 Models by Daily Token Volume (June 6, 2026)
| Rank | Model | Company | Origin | Daily Tokens |
|---|---|---|---|---|
| 1 | DeepSeek V4 Flash | DeepSeek | 🇨🇳 | 619B |
| 2 | Hy3 Preview | Tencent | 🇨🇳 | 451B |
| 3 | MiniMax M3 | MiniMax | 🇨🇳 | 447B |
| 4 | MiMo-V2.5 | Xiaomi | 🇨🇳 | 327B |
| 5 | DeepSeek V4 Pro | DeepSeek | 🇨🇳 | 300B |
| 6 | Claude Sonnet 4.6 | Anthropic | 🇺🇸 | 263B |
| 7 | Claude Opus 4.8 | Anthropic | 🇺🇸 | 178B |
| 8 | DeepSeek V3.2 | DeepSeek | 🇨🇳 | 161B |
| 9 | Owl Alpha | OpenRouter | — | 158B |
| 10 | Gemini 3 Flash Preview | 🇺🇸 | 156B |
Source: OpenRouter.ai/rankings, Today view, June 6, 2026.
DeepSeek holds three positions in the top 10 — ranks 1, 5, and 8. Combined, its three models processed 1.08 trillion tokens in a single day. No other company comes close at the model level.
Anthropic places two models in the top 10 — Sonnet at #6 and Opus at #7 — for a combined 441 billion daily tokens. Both in the top 10 simultaneously is a meaningful indicator of broad enterprise adoption across different price points.
Google's Gemini 3 Flash Preview at #10 is the only Google model in today's top 10, but the company's 12.5% weekly company share (third overall) reflects a wider footprint across specialized models not captured in this daily snapshot.
The DeepSeek Story: Three Models, One Dominant Company
DeepSeek's position deserves attention beyond the headline number. It is the only company running a three-model strategy that spans the full stack: a Flash variant for high-throughput low-latency tasks (#1 daily, 619B tokens), a Pro variant for complex reasoning (#5, 300B), and a legacy V3.2 that still drives significant traffic (#8, 161B).
This breadth is strategically significant. A developer building an application on DeepSeek can route simple tasks to V4 Flash and complex tasks to V4 Pro within the same provider relationship. No switching costs, no vendor mixing. DeepSeek has built what amounts to a full-stack inference offering — and the token volumes show developers have noticed.
At 17.6% weekly platform share, DeepSeek's token volume is larger than Google's (12.5%) and OpenAI's (8.4%) combined.
The Use Case Split: Programming Is Where the Real Battle Is
OpenRouter breaks down token volume by use case. The programming category is the most revealing.
Top Models for Programming (Weekly Token Volume)
| Rank | Model | Company | Weekly Tokens | Share |
|---|---|---|---|---|
| 1 | MiMo-V2.5 | Xiaomi 🇨🇳 | 1.68T | 18.0% |
| 2 | MiniMax M3 | MiniMax 🇨🇳 | 1.04T | 11.1% |
| 3 | DeepSeek V4 Flash | DeepSeek 🇨🇳 | 624B | 6.7% |
| 4 | Hy3 Preview | Tencent 🇨🇳 | 559B | 6.0% |
| 5 | DeepSeek V4 Pro | DeepSeek 🇨🇳 | 476B | 5.1% |
| 6 | MiMo-V2.5-Pro | Xiaomi 🇨🇳 | 394B | 4.2% |
| 7 | Step 3.7 Flash | StepFun 🇨🇳 | 352B | 3.8% |
| 8 | Claude Opus 4.7 | Anthropic 🇺🇸 | 333B | 3.6% |
| 9 | Claude Opus 4.8 | Anthropic 🇺🇸 | 317B | 3.4% |
Source: OpenRouter.ai/rankings, Programming category, June 2026.
Xiaomi's MiMo-V2.5 holds 18% of all programming tokens on the platform — nearly one in five coding tokens routes through a Xiaomi model. When you add MiMo-V2.5-Pro at #6 (4.2%), Xiaomi's two MiMo models together represent over 22% of all coding traffic on OpenRouter.
This matters for investors because coding is not a niche use case — it is the largest single use case on the platform. Whoever wins coding wins the largest piece of the AI inference market.
Anthropic places two models in coding's top 9 (Opus 4.7 at 3.6%, Opus 4.8 at 3.4%). Combined at ~7%, Anthropic maintains a meaningful coding presence despite the volume dominance of Chinese models.
Agentic AI: The Fastest-Growing Volume Driver
The top apps using OpenRouter tell you where the volume growth is coming from.
Top Apps by Daily Token Volume (June 6, 2026)
| Rank | App | Daily Tokens |
|---|---|---|
| 1 | Hermes Agent (Nous Research) | 669B |
| 2 | Kilo Code | 175B |
| 3 | OpenClaw | 160B |
| 4 | Claude Code (Anthropic) | 96.6B |
| 5 | pi | 69B |
Source: OpenRouter.ai/rankings, Top Apps view, June 6, 2026.
Hermes Agent alone processes 669 billion tokens per day — more than Claude Sonnet 4.6 and Claude Opus 4.8 combined. It is an open-source autonomous AI agent that runs persistently with memory across sessions. This is not a chatbot. This is a system running continuously, consuming tokens at a rate that would have been unimaginable for a single application two years ago.
Kilo Code (175B) and OpenClaw (160B) are coding and automation agents. Four of the top five apps are autonomous agents or agentic coding tools. The implication for token volume is structural: a user running an agentic workflow consumes orders of magnitude more tokens per session than a user having a conversation.
Tool call volume confirms the trend. The models leading on tool calls — the building block of agentic workflows — are:
| Rank | Model | Daily Tool Calls | Share |
|---|---|---|---|
| 1 | Hy3 Preview (Tencent) | 33M | 8.9% |
| 2 | DeepSeek V4 Flash | 32M | 8.6% |
| 3 | Gemini 3 Flash Preview | 21.4M | 5.7% |
| 4 | Claude Sonnet 4.6 | 19.5M | 5.2% |
| 5 | MiniMax M3 | 19.3M | 5.2% |
Tencent's Hy3 leads all models in tool call volume. For investors in Tencent, this is the most important data point in the entire OpenRouter rankings: Hy3 is not just a high-volume model — it is the top choice for developers building autonomous AI systems.
The Intelligence Benchmark: Quality Still Has a Champion
Token volume is not the only scoreboard. OpenRouter also publishes an Intelligence Index ranking — a composite benchmark of model capability.
Intelligence Index (Artificial Analysis, via OpenRouter)
| Rank | Model | Company | Score |
|---|---|---|---|
| 1 | Claude Opus 4.8 (Adaptive Reasoning, Max Effort) | Anthropic | 61.4 |
| 2 | GPT-5.5 (xhigh) | OpenAI | 60.2 |
| 3 | Claude Opus 4.7 (Adaptive Reasoning, Max Effort) | Anthropic | 57.3 |
| 4 | Gemini 3.1 Pro Preview | 57.2 | |
| 5 | GPT-5.4 (xhigh) | OpenAI | 56.8 |
| 6 | Qwen3.7 Max | Qwen/Alibaba | 56.6 |
| 7 | Gemini 3.5 Flash (high) | 55.3 | |
| 8 | MiniMax-M3 | MiniMax | 54.7 |
| 9 | Kimi K2.6 | Moonshot AI | 53.9 |
| 10 | MiMo-V2.5-Pro | Xiaomi | 53.8 |
Source: OpenRouter.ai/rankings, Benchmarks view, June 2026.
Claude Opus 4.8 is the #1 model by intelligence despite ranking #7 by daily token volume. This gap — the largest in the rankings — tells you exactly what kind of work developers are routing to Anthropic: high-stakes reasoning tasks where quality matters more than throughput. Legal analysis, complex code review, financial modeling, enterprise document processing.
GPT-5.5 (OpenAI) ranks #2 at 60.2 — and notably, GPT-5.5 appears in the intelligence rankings while GPT-5.4 appears in daily usage. OpenAI is running a generational model transition in real time.
The most striking intelligence benchmark result for investors: MiMo-V2.5-Pro (Xiaomi) ranks #10 at 53.8, while dominating coding volume. A model that ranks 10th on intelligence is #1 on programming. That is the strongest possible evidence that coding is increasingly a throughput problem, not a reasoning problem — and that whoever builds the most efficient inference pipeline for coding tasks will capture the largest single use case on the platform.
Winners and Losers: The Investor Read
WINNER: Nvidia (NVDA)
Every model in every ranking runs on GPU infrastructure. DeepSeek's three top-10 models, Xiaomi's two MiMo variants, Tencent's Hy3, Anthropic's two Opus/Sonnet models — all of them consume GPU compute at scale. The ~29 trillion weekly tokens shown in OpenRouter's market share data represent raw GPU demand, and that demand is growing on the chart's trajectory from 9T to over 27T over the past year.
Investable angle: NVDA is the infrastructure bet that wins regardless of which model leads next week.
WINNER: Tencent (700.HK)
Hy3 Preview is Tencent's breakout model and the data is unambiguous: #2 by daily tokens (451B), #1 by tool call volume (33M, 8.9%), strong in Python (#2 at 8.2%), and a consistent top-4 across use cases. Hy3 is not a volume model — it is the leading model for agentic AI workflows, which are the fastest-growing token consumption category on the platform.
Tencent is investable via 700.HK (Hong Kong) or OTC ADRs. The agentic AI use case is the highest-growth segment on OpenRouter and Tencent owns it today.
Investable angle: Tencent is the agentic AI bet. Hy3's #1 tool call position puts it at the center of the platform's fastest-growing use case.
WINNER: Xiaomi (1810.HK)
MiMo-V2.5 owns 18% of all coding tokens on OpenRouter — nearly one in five. With MiMo-V2.5-Pro adding another 4.2%, Xiaomi controls over 22% of the programming use case. Coding is the single largest category on the platform.
Xiaomi is traditionally known as a hardware company. The OpenRouter data shows it has built — quietly — the most-used coding AI model in the world by token volume.
Investable angle: Xiaomi (1810.HK) offers direct exposure to the dominant coding model. If programming remains the largest AI use case, MiMo's position has real revenue implications for Xiaomi's cloud/API business.
NUANCED: Anthropic (private — watch AMZN, GOOGL)
Anthropic's position is the most counterintuitive in the rankings. It ranks #2 by weekly company tokens (14.8%) — higher than Google (12.5%), OpenAI (8.4%), and every individual Chinese company except DeepSeek. It holds two models in today's top 10 simultaneously. And it holds the #1 and #3 intelligence benchmark scores.
The bearish read: Chinese models dominate volume, coding, and tool calls. Anthropic's 14.8% is achieved at the high end of the quality spectrum — any erosion of the quality premium compresses its position.
The bullish read: 14.8% of a ~29T weekly token platform is not a niche. Anthropic is embedded across enterprise workflows at a scale that pure volume metrics obscure. The intelligence gap — Claude Opus 4.8 at 61.4 vs. the nearest Chinese model (MiMo-V2.5-Pro at 53.8) — is 7.6 points, not trivial.
Anthropic is private, but Amazon (AMZN) is its primary cloud partner and major investor. Google (GOOGL) is also an investor and ranks #3 by weekly company tokens independently.
WATCH: OpenAI / Microsoft (MSFT)
OpenAI ranks #4 by company at 8.4% (2.46T weekly tokens). GPT-5.5 ranks #2 on the intelligence benchmark at 60.2. But the daily model leaderboard tells a more sobering story: no OpenAI model appears in today's top 10 by individual model volume. The company's weekly share is distributed across its model family rather than concentrated in a breakout hit.
Microsoft's investment in OpenAI is tied to platform share that is real but not growing as fast as the Chinese alternatives. Enterprise and Copilot integrations — which do not appear in OpenRouter's developer-API data — are the Microsoft bull case, and they are not measured here.
Investor watch: Track whether any OpenAI model cracks the daily top 10. It would signal a shift in developer preference that the weekly company share does not yet show.
The Big Picture: What the Rankings Actually Tell You
Three structural shifts are visible in the June 2026 OpenRouter data:
1. DeepSeek is the AI company most investors aren't talking about. At 17.6% of weekly platform tokens — more than Google and OpenAI combined — DeepSeek is the largest single AI provider by developer token usage. It is private and based in China, which limits direct investment options, but it sets the competitive benchmark every other company is racing against.
2. Programming and agentic AI are consuming the platform. MiMo-V2.5 owns 18% of all coding tokens. Hermes Agent alone processes 669B tokens per day. The models and companies winning these two use cases — Xiaomi (coding) and Tencent (agentic/tool calls) — are the real volume winners of the current cycle.
3. Quality and volume have completely diverged. Claude Opus 4.8 leads every intelligence benchmark by a meaningful margin while ranking 7th by daily usage. GPT-5.5 is #2 on intelligence while absent from the daily model top 10. The market has separated: volume is won by optimizing for throughput and cost, quality is won by optimizing for reasoning. Both businesses can exist simultaneously, but they are no longer the same business.
How to Track This in Real Time
OpenRouter's rankings update daily at openrouter.ai/rankings. For investors, the key views to check weekly:
- Market Share → by company: The weekly company view is more stable than daily model rankings and shows structural position
- Categories → Programming: The single largest use case; whoever leads here leads the platform
- Tool Calls: The leading indicator for agentic AI adoption, the fastest-growing segment
- Benchmarks: Tracks whether the intelligence gap between US and Chinese models is narrowing
For the publicly traded stocks most directly exposed to this data — NVDA, TCEHY (Tencent OTC), XIACF (Xiaomi OTC), AMZN, GOOGL, MSFT — set real-time price alerts so you can track how the market is pricing these AI leadership shifts as the rankings evolve week to week.
Set real-time price alerts on NVDA, AMZN, GOOGL, and MSFT in Stock Alarm Pro to track how OpenRouter's AI race data translates into stock price moves as the rankings change.
Source
All data in this article is sourced exclusively from OpenRouter.ai/rankings, June 6, 2026. Figures include: Today (daily model leaderboard), Market Share (weekly by company), Programming category (weekly), Tool Calls (daily), Benchmarks (Intelligence Index via Artificial Analysis), and Top Apps (daily). No estimates or third-party projections are used. All numbers are taken directly from the OpenRouter rankings interface.
