OpenRouter Token Data: A Clean Signal of Developer AI Adoption
OpenRouter’s AI model rankings are based on token consumption across thousands of live applications, providing a neutral, usage-driven measure of developer AI adoption and enterprise AI usage that cuts through benchmark hype and marketing claims by showing where real workloads run and how traffic is routed between competing models in production. The June AI model rankings show large-scale consolidation: the top ten companies process roughly 19 trillion tokens, with the top four alone handling more than half of all activity. That concentration turns OpenRouter into a kind of de facto market share chart for foundation models. DeepSeek leads with 3.1 trillion tokens and 16.3% share, ahead of Anthropic’s 2.94 trillion tokens and 15.5% share, highlighting how cost, performance, and open-weights availability are reshaping routing choices at scale.
Google’s Nano Banana Sweep in the Image Generation Market
In image generation, OpenRouter popular models data points to a far less fragmented field: Google holds a commanding lead. Its three Nano Banana models, built on Gemini image backends, account for about 89% of all image generation traffic on the platform, making Google the clear image generation market leader among OpenRouter users. Nano Banana (Gemini 2.5 Flash Image) leads with 1.71 million requests and a 40.7% share, followed by Nano Banana 2 at 1.21 million requests and 28.8%, and Nano Banana Pro at 825,000 requests and 19.6%. The lineup splits by use case: Nano Banana for high-volume, speed-sensitive jobs, Nano Banana 2 for higher-quality text-to-image at a lower price point than Pro, and Nano Banana Pro for high-fidelity commercial work with SynthID watermarking and integration into Google Ads and Workspace.
Claude, GPT-4 and the Battle for Text-Based Workloads
The text-focused AI model rankings for June show a more competitive landscape. On the company side, Anthropic sits second overall on OpenRouter with 15.5% of token volume, driven by Claude models that appeal to enterprises willing to pay for strong reasoning, coding, and agentic behavior. At the model level, Claude Opus 4.7 and Claude Sonnet 4.6 sit near the top of the leaderboard, with Opus 4.7 emerging as the highest-ranked closed-source model and a reference point for complex production deployments. According to OpenRouter’s model rankings, Opus 4.7 leads GPT-5.4 and Gemini 3.1 Pro on key agentic benchmarks such as SWE-bench Pro and SWE-bench Verified, and Anthropic’s data links Claude Code to about 4% of all public GitHub commits. GPT-family models and other proprietary systems remain heavily used, but developer routing choices show no single winner for text yet.
Open-Weights Surge and Enterprise vs. Consumer Adoption Patterns
Below the image generation headline, the most striking shift in developer AI adoption is the rise of open-weights leaders. DeepSeek V4 Flash tops the overall model chart with 10.9 trillion tokens and nearly 10x month-on-month growth, driven by a sparse MoE design, 1M-context support, and aggressive pricing that makes it attractive for high-volume agentic and automation workloads. Tencent’s Hy3 Preview is an even sharper newcomer story, going from zero to near parity with V4 Flash at 10.7 trillion tokens in a single month, thanks to its 295B-parameter MoE design, long-context handling, and low per-token cost. Enterprises appear to favor Claude Opus 4.7 and Sonnet 4.6 when accuracy and reliability matter most, while consumer and cost-sensitive apps push massive traffic through DeepSeek and Hy3, despite risks like V4 Flash’s tendency to hallucinate when it lacks an answer.
Monthly Leaderboards as a Forward Indicator of AI Market Structure
OpenRouter’s monthly leaderboard has become a moving snapshot of the AI market’s structure, showing where developers send traffic as new models launch and capabilities shift. June’s AI model rankings show clear consolidation: the top four companies by token volume already process most of the platform’s 19 trillion tokens, and that share is growing as DeepSeek and Tencent expand. At the same time, the image leaderboard displays near-total dominance by Google’s Nano Banana trio, suggesting that image generation may be consolidating faster than text or agents. For enterprises planning long-term architecture, these numbers matter more than isolated benchmarks. They reveal which systems are reliable enough for production, which open-source stacks are catching up within a four-month capability lag, and how quickly routing patterns can swing when a new high-performing, lower-cost model like Hy3 Preview appears in the catalog.
