Data · Which AI do businesses actually run? · Week of October 26 to November 1, 2026
There is no single best model, and the market splits two ways. In the week of October 26 to November 1, 2026, DeepSeek models processed 36.9% of all tokens on OpenRouter, the most of any maker, while Claude Opus 5.5 from Anthropic took the largest share of spend on running business workflows (24.1% over the last 30 days). Source: OpenRouter (openrouter.ai/rankings), as of November 2, 2026. Licensed under CC BY 4.0.
| Job | First | Second | Third |
|---|---|---|---|
| Running multi-step workflows 20.3% of all spend | Claude Opus 5.5 Anthropic · 24.1% | GPT 6 Astra OpenAI · 6.4% | Kimi K3 Moonshot AI · 6.0% |
| Sorting and tagging records 11.8% of all spend | Claude Opus 5.5 Anthropic · 11.0% | GPT 6.1 Sol OpenAI · 6.2% | GPT 5.6 Sol OpenAI · 5.5% |
| Pulling data out of documents 7.2% of all spend | Claude Opus 5.5 Anthropic · 11.3% | Claude Sonnet 5.5 Anthropic · 4.8% | Gemini 3.8 Flash Google · 4.4% |
| Reformatting and cleaning data 3.1% of all spend | GLM 5.3 Flash Z.ai · 13.5% | Claude Opus 5.5 Anthropic · 13.5% | Claude Sonnet 5.5 Anthropic · 5.0% |
| Summarizing 1.2% of all spend | Claude Opus 5.5 Anthropic · 13.9% | GPT 6 Astra OpenAI · 6.0% | Claude Sonnet 5.5 Anthropic · 4.8% |
| Writing and drafting 3.7% of all spend | Claude Opus 5.5 Anthropic · 11.9% | Claude Sonnet 5.5 Anthropic · 8.7% | GLM 5.3 Flash Z.ai · 4.6% |
| Customer support replies 1.2% of all spend | Gemini 3 Flash Preview Google · 6.3% | Claude Opus 5.5 Anthropic · 6.3% | Claude Sonnet 5.5 Anthropic · 6.1% |
| Research reports 1.4% of all spend | Claude Opus 5.5 Anthropic · 21.0% | Kimi K3 Moonshot AI · 8.3% | Claude Sonnet 5.5 Anthropic · 5.2% |
| Answering questions from knowledge 3.5% of all spend | Claude Opus 5.5 Anthropic · 14.6% | Kimi K3 Moonshot AI · 12.5% | GPT 6.1 Sol OpenAI · 4.6% |
| Translation 0.4% of all spend | GPT 5.6 Luna OpenAI · 6.0% | GPT 6 Astra OpenAI · 5.7% | DeepSeek V4.1 Flash DeepSeek · 5.4% |
| Finance work 0.4% of all spend | Claude Opus 5.5 Anthropic · 14.1% | Claude Fable 5.1 Anthropic · 5.6% | Kimi K3 Moonshot AI · 5.0% |
| Database queries 0.5% of all spend | Claude Fable 5.1 Anthropic · 19.7% | Claude Opus 5.5 Anthropic · 16.7% | Claude Sonnet 5.5 Anthropic · 7.6% |
What it means
DeepSeek leads on volume: more text runs through its models than any other maker's. Spend tells a different story. Anthropic models lead spend on 9 of the 12 business jobs in the table, so the models that process the most text are not the ones businesses pay the most to use for this work. The exceptions are reformatting and cleaning data (GLM 5.3 Flash), customer support replies (Gemini 3 Flash Preview), translation (GPT 5.6 Luna). For a small business the practical read is to match the model to the job: start with the models businesses already pay for on work that touches customers, money or records, and test a high-volume model on sorting, tagging and bulk drafting that a person reviews.
| Maker | Week of Oct 26 | Week of Sep 28 |
|---|---|---|
| DeepSeek | 36.9% | 22.3% |
| OpenAI | 11.5% | 11.4% |
| Z.ai | 10.6% | 8.7% |
| Anthropic | 7.4% | 3.9% |
| Tencent | 6.3% | 5.2% |
| MiniMax | 3.5% | Not in top 9 |
| StepFun | 3.5% | Not in top 9 |
| NVIDIA | 3.5% | 4.0% |
| 3.4% | 4.1% |
| Model | Maker | Share of tokens |
|---|---|---|
| DeepSeek V4.1 Flash | DeepSeek | 24.8% |
| GLM 5.3 Flash | Z.ai | 7.8% |
| DeepSeek V4.1 Flash | DeepSeek | 7.7% |
| GPT 6 Luna | OpenAI | 5.0% |
| HY4 Preview | Tencent | 4.7% |
| Claude Opus 5.5 | Anthropic | 3.5% |
| Step 5 Preview | StepFun | 3.2% |
| MiniMax M3.1 Flash | MiniMax | 3.2% |
| GPT 5.6 Luna | OpenAI | 2.2% |
| All other models | 37.9% |
OpenRouter is a service developers and companies use to reach many AI models through one connection, so its rankings show what builders actually run rather than what vendors claim. It is developer and API traffic through one service, not all business use of AI: people using ChatGPT, Claude or Copilot directly do not appear here.
Token share is each maker's or model's share of all tokens processed in a complete week (Monday to Sunday). OpenRouter shows the nine largest makers and models by name and groups the rest. Spend by job uses OpenRouter's classification of requests into task types over the last 30 days; the share is each model's share of spend on that task. A high share means businesses chose to pay for that model for that job, not that it scored best on a benchmark.
Source: OpenRouter (openrouter.ai/rankings), as of November 2, 2026. Licensed under CC BY 4.0. Refreshes weekly. Download the data (JSON).
Not sure which model fits your workflow? Fifteen minutes to match the job to the right tool.