What OpenRouter Usage Data Reveals About AI Model Adoption
OpenRouter usage data is a live record of AI model adoption trends, showing which providers win real developer traffic, how production workloads shift, and where market power concentrates across text and image models. As a neutral routing layer, OpenRouter sits between applications and dozens of AI providers, letting developers switch models without vendor lock-in. That means token counts and request volumes reflect deliberate choices, not captive usage. In the latest figures, the top ten companies processed about 19 trillion tokens, with the top four handling more than half of all activity. DeepSeek leads with 3.1 trillion tokens, Anthropic follows with 2.94 trillion, Google is third at 2.51 trillion, and OpenAI sits fourth with 1.65 trillion. These numbers show a market where a few providers capture most traffic, and where cost, capability, and integration now matter more than brand alone.
Google’s Nano Banana Models Dominate Image Generation Market Share
Google has taken a commanding lead in AI image generation market share on OpenRouter, powered by its Nano Banana lineup under the Gemini family. Nano Banana (Gemini 2.5 Flash Image) logs 1.71 million requests and a 40.7% share, Nano Banana 2 (Gemini 3.1 Flash Image) reaches 1.21 million requests and 28.8%, and Nano Banana Pro (Gemini 3 Pro Image) adds 825,000 requests at 19.6%. Together, these three models account for about 89% of all image generation traffic on the platform. Their success began with an unplanned viral moment: a late-night upload to LMArena that turned a placeholder name into a product brand. From there, Google split the line into speed, quality, and high-fidelity tiers, integrating Nano Banana Pro into Google Ads and Workspace and backing it with SynthID watermarking, aligning developer AI preferences with enterprise-grade authenticity needs.
Text Model Leaders: DeepSeek’s Cost Edge and Anthropic’s Enterprise Pull
Beyond images, OpenRouter’s token volumes show which text and multimodal models anchor production systems. DeepSeek stands out with 16.3% of token volume, built on its R1 and V4-Pro models that aim for frontier-level performance at lower cost. V4-Pro scores 1554 on the GDPval-AA agentic benchmark and, on the Artificial Analysis Intelligence Index, costs USD 1,071 (approx. RM4,600) versus USD 4,811 (approx. RM20,600) for Claude Opus 4.7, a gap large enough to reshape procurement decisions for companies running billions of calls. Anthropic holds 15.5%, with Claude still a go-to choice for complex reasoning and coding tasks where enterprises accept a premium for capability and support. Google’s 13.2% share reflects the reach of Gemini across its product stack, while OpenAI’s 8.7% slice marks a shift from default status to one option among several, especially in price-sensitive, API-heavy workloads.
Rising Challengers and the Long Tail in Image and Text Models
OpenRouter’s marketplace also highlights a long tail of challengers gaining meaningful, if smaller, shares. In image models, GPT-5.4 Image 2 holds 2.8% of requests, showing OpenAI’s steady but secondary position behind Google in this category. Seedream 4.5 by ByteDance-Seed reaches 2.4%, and Grok Imagine Image Quality from xAI records 2.2%, indicating that developers still explore alternatives when they offer specific strengths or ecosystem ties. On the text side, Xiaomi’s 8.6% token share shows how device ecosystems can push substantial traffic through shared infrastructure. Tencent and other large technology players also appear among the top ten, reinforcing that AI model adoption trends are not limited to a single set of incumbents. For enterprises, this long tail represents optionality and experimentation, even as most production volume clusters around a few dominant providers.
Market Consolidation and What It Means for Enterprise AI Choices
The combined picture from OpenRouter usage data is of a consolidating market where a handful of players capture most production workloads across text and image generation. The top four companies on the platform already account for more than half of all token volume, and in image generation a single vendor controls nearly 90% of traffic. For enterprises, this concentration cuts both ways: it offers clear, mature options with proven scale, but increases the risk of dependency on a small set of providers. Neutral routing layers like OpenRouter help counter this by making it easier to rebalance workloads across models as prices, capabilities, or compliance needs change. In practice, procurement teams now have to weigh not only raw performance, but also ecosystem lock-in, benchmark-verified cost differences, and the cultural pull that can push developer AI preferences toward a specific brand, as Nano Banana’s rise demonstrates.
