Top 10 Large Language Models by OpenRouter Usage in October 2026

Which large language models are attracting usage? This snapshot uses OpenRouter’s weekly token-usage leaderboard, not global LLM market share. It ranks individual model entries rather than companies. The captured rankings page labels its usage data through October 28, 2026.

OpenRouter says models are ranked by tokens processed through its API. We reproduce the weekly table’s order and displayed token notation without calculating global percentages or estimating missing traffic. The figures are rounded as displayed, so they should not be treated as exact raw counts. This is a platform-specific usage proxy, not a survey of every provider.

The leading model entries

1. DeepSeek V4.1 Flash 1017. OpenRouter lists DeepSeek V4.1 Flash 1017 as a model entry. The weekly table shows 39.6T tokens for DeepSeek V4.1 Flash 1017. See its model page. Keep this dated variant separate from the similarly named model below. Combining releases would produce a different ranking from the source.
2. DeepSeek V4.1 Flash. OpenRouter lists DeepSeek V4.1 Flash as a model entry. The weekly table shows 16.6T tokens for DeepSeek V4.1 Flash. See its model page. The table treats this as its own model entry. Our recommendation is to preserve that identity when making historical comparisons.

The next platform-usage positions

1. GLM 5.3 Flash. OpenRouter lists GLM 5.3 Flash as a model entry. The weekly table shows 15T tokens for GLM 5.3 Flash. See its model page. Our interpretation concerns traffic through this platform, not sales or adoption across every application using language models.
2. GPT-6 Luna. OpenRouter lists GPT-6 Luna as a model entry. The weekly table shows 9.59T tokens for GPT-6 Luna. See its model page. A high token volume is not a controlled quality benchmark. We would evaluate its actual suitability separately on the intended task.

Preview and assistant entries

1. Hy4 preview. OpenRouter lists Hy4 preview as a model entry. The weekly table shows 8.96T tokens for Hy4 preview. See its model page. The preview label belongs in the name. Do not silently treat a preview entry as an interchangeable production release.
2. Claude Opus 5.5. OpenRouter lists Claude Opus 5.5 as a model entry. The weekly table shows 7.86T tokens for Claude Opus 5.5. See its model page. For procurement, our advice is to compare task results and operating conditions rather than turn observed traffic into a recommendation.

Additional developer choices

1. Step 5 Preview. OpenRouter lists Step 5 Preview as a model entry. The weekly table shows 5.39T tokens for Step 5 Preview. See its model page. Its place here describes the captured weekly table. It does not establish an enduring position or predict next year’s demand.
2. GLM 5.3. OpenRouter lists GLM 5.3 as a model entry. The weekly table shows 5.15T tokens for GLM 5.3. See its model page. This is distinct from the Flash entry above. Grouping models by their developer would change the unit of analysis.

The final entries in this snapshot

1. Nemotron 3 Ultra (free). OpenRouter lists Nemotron 3 Ultra (free) as a model entry. The weekly table shows 4.44T tokens for Nemotron 3 Ultra (free). See its model page. Preserve the free-endpoint label when interpreting the list. This snapshot is not a revenue ranking or a paid-customer count.
2. MiMo-V2.6-Flash. OpenRouter lists MiMo-V2.6-Flash as a model entry. The weekly table shows 4.1T tokens for MiMo-V2.6-Flash. See its model page. The lower position in this selection does not itself show inferior capability. The ranking measures usage, not a product test.

OpenRouter’s Data API documentation defines its daily token totals as prompt tokens plus completion tokens. That definition helps explain the measure; it does not convert this weekly display into a global denominator. Consult the captured ranking’s source and dataset documentation before making comparisons. We have not computed market-share percentages.

Related reading: model ownership costs; batch AI procurement; AI infrastructure; AI memory portability; healthcare AI evidence.