OpenRouter has become one of the most useful barometers for real-world AI adoption — a neutral marketplace where developers and enterprises route traffic across dozens of models, with no vendor lock-in. The token volumes it generates tell you something that marketing slides can’t: which AI companies are actually getting used.
The latest data paints a striking picture. Combined, the top 10 companies on the platform processed roughly 19 trillion tokens, with the top four alone accounting for more than half of all activity. Here’s how it breaks down.
1. DeepSeek — 3.1T tokens (16.3%)
DeepSeek sits at the top, and it isn’t close. The Chinese lab — founded in 2023 and backed by quantitative hedge fund High-Flyer — commands 16.3% of all token volume on OpenRouter, more than any other single provider.
The number reflects a momentum that’s been building for over a year. When DeepSeek released R1 in early 2025, it stunned the industry with benchmark performance comparable to OpenAI’s top models at a fraction of the cost. Its subsequent releases only extended the lead. DeepSeek V4-Pro now scores 1554 on GDPval-AA, making it the leading open-weights model on agentic benchmarks — the category that matters most for production deployments.
The cost advantage is just as sharp. Running the Artificial Analysis Intelligence Index benchmark, V4-Pro costs $1,071 compared to $4,811 for Claude Opus 4.7 — more than 4x cheaper. For companies running billions of model calls per month, that math is impossible to ignore.
The migration is already happening. Flo Crivello, CEO of AI agent platform Lindy, switched his company entirely to DeepSeek V4, citing millions of dollars in savings and an actual improvement in performance on core use cases. Lindy is not a hobbyist project — it’s a funded, production-grade platform, and the move was the outcome of months of systematic benchmarking.
2. Anthropic — 2.94T tokens (15.5%)
Anthropic holds second place with 15.5% of token volume — a position that speaks to the genuine enterprise pull its Claude models have built.
The company has had a remarkable run: revenue grew from $100 million in 2023 to $4.5 billion by mid-2025, and its valuation has since climbed to $965 billion. On OpenRouter, Claude models attract developers who need the best available capabilities and are willing to pay the premium — particularly for complex reasoning, coding, and agentic tasks.
The competitive pressure, however, is real. Anthropic’s API-dependent business model is the most price-sensitive segment of the market, and the gap between its models and top open-source alternatives is narrowing. Open-source models currently trail the frontier by approximately four months — a gap that can close quickly if capability progress slows. Staying in second place on OpenRouter while DeepSeek grows above it will be one of the defining tests for the company.
3. Google — 2.51T tokens (13.2%)
Google is third with 13.2% of volume, driven primarily by the Gemini family. The company has a structural advantage that neither DeepSeek nor Anthropic can easily replicate: its models are embedded in products that billions of people use every day — Search, Android, Workspace — which generates usage whether or not developers consciously choose Gemini.
On OpenRouter specifically, Google’s strong showing reflects the Gemini 2.x releases, which have been well-received for their long-context handling and multimodal capabilities. The combination of performance and distribution makes Google one of the harder players to displace.
4. OpenAI — 1.65T tokens (8.7%)
OpenAI sits fourth — a placement that would have seemed unthinkable two years ago when GPT-4 was the unambiguous default for every serious AI application. The company still holds the strongest consumer brand in the space (ChatGPT has nearly a billion daily users) but on OpenRouter — a platform where developers route traffic deliberately — it now ranks below three competitors.
The gap between OpenAI’s 8.7% share and DeepSeek’s 16.3% is a signal worth paying attention to. Developer-facing usage is where pricing pressure shows up first, and it is showing up.
5. Xiaomi — 1.64T tokens (8.6%)
Xiaomi’s appearance in fifth place is the most surprising entry on the list. The Chinese consumer electronics giant — better known for smartphones, smart home devices, and EVs — has been quietly building out AI model capabilities and appears to be routing significant traffic through OpenRouter. Its 8.6% share, nearly identical to OpenAI’s, reflects both the breadth of its device ecosystem and China’s aggressive push into AI infrastructure across non-traditional players.
6–9: The Challengers
Tencent (6th, 7.8%) brings the weight of one of the world’s largest technology conglomerates to the table. Its AI model efforts, while less covered in Western press than DeepSeek’s, are clearly generating substantial real-world usage.
MiniMax (7th, 7.2%) is a Shanghai-based startup that has built a reputation for long-context and multimodal capabilities. Chinese models including MiniMax surged on OpenRouter during OpenClaw mania earlier this year, when four of the top five models by call volume were Chinese.
OpenRouter itself appears at 8th with 5.9% — reflecting usage of its own native or aggregated model offerings on the platform.
Qwen (9th, 4.7%), Alibaba’s model family, rounds out the named providers. Qwen has been one of the stronger open-weight model series globally, with Qwen3 variants earning recognition on capability benchmarks. Its presence on OpenRouter has been growing steadily.
The Bigger Picture
The most striking takeaway from this data isn’t any single company’s number — it’s the geographic distribution. Chinese providers (DeepSeek, Xiaomi, Tencent, MiniMax, Qwen) collectively account for roughly 44% of token volume across the top 10. That is not a China-specific story anymore; it’s a global developer story. Developers worldwide are choosing Chinese models because they perform well and cost less.
OpenRouter’s own historical data shows that proprietary models maintained approximately 70% of total token volume through 2025, with open-source steady at 30% — but within that open-source share, Chinese models displaced rest-of-world alternatives dramatically. The trend has continued into 2026.
For enterprises and developers still building on a single incumbent provider, the OpenRouter data is a useful reality check. The market has fragmented. The performance gap between frontier proprietary models and top open-source alternatives has narrowed to months, not years. And the cost differential in many cases exceeds 4x.
That doesn’t mean Anthropic or OpenAI are in trouble — enterprise relationships, compliance infrastructure, and developer trust are real moats. But the rankings on a neutral routing platform like OpenRouter are among the most honest signals the industry produces, and right now they show a market in the middle of a significant reordering.