# DeepSeek took 53% of OpenRouter's 27T daily tokens, a post reports

> A post reports DeepSeek accounted for 53% of all tokens on OpenRouter in one day, with open-weight labs together above 75%.

- Status: reported (One outlet, not independently verified yet.)
- Topic: Models
- Published: 2026-11-06T10:36:15+00:00
- Updated: 2026-11-06T10:36:15+00:00
- Web page: https://ai-news.ikangai.com/story/deepseek-took-53-of-openrouter-s-27t-daily-tokens-a-post-reports/
- Seen in: Latent Space
- Change: other (Other)

DeepSeek accounted for 53% of all tokens routed through OpenRouter on a single day, according to a Reddit post in the LocalLLaMA community, which put that day's total at about 27 trillion tokens. The same post says open-weight labs together made up more than 75% of the volume. OpenRouter is a gateway that forwards requests to many backing providers. The share therefore reflects which models developers choose to send through that one service, not total AI usage across the market.

Commenters on the post raised the most obvious limit of that reading. They argued that OpenRouter's users are already more likely to pick open-weight models, while large organizations often buy directly from vendors, so the platform may not represent enterprise spending. One commenter cited public estimates that Google processes around 100 trillion tokens a day, roughly three times OpenRouter's 27 trillion. Inferact, a vLLM company, separately puts OpenRouter's average at 19 trillion tokens a day of open-weight traffic.

The more technical objection concerns cost on agent workloads. Commenters claimed OpenRouter keeps a prompt pinned to the same provider for only about 10 minutes, and that many backing providers retain their cache for a short time. Prompt caching stores the attention state for a repeated prefix so the model doesn't have to reprocess it on each call. If the cache expires, a long agent history gets recomputed repeatedly, and the user pays for that work again. One commenter recommended calling DeepSeek's own API directly, where they said cache retention lasts "up to days" instead of minutes. Those are the commenters' claims about provider behavior, and the post itself doesn't verify them.

One commenter reported that a day-long background agent workflow cost about $4 on DeepSeek, compared with over $400 for a similar session on Claude. The commenter called this anecdotal, and it comes from one session, not a measured comparison. Another commenter cited opencode.ai/data showing DeepSeek at 84% of tokens with 24 trillion tracked, which points to a coding tool's traffic rather than OpenRouter's.

## Why it matters

Usage on a routing platform signals developer preference, but companies buying AI at scale should not treat it as a proxy for their own cost or model choice.

---

AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it.

Cite as: IKANGAI AI News, "DeepSeek took 53% of OpenRouter's 27T daily tokens, a post reports", 2026-11-06, https://ai-news.ikangai.com/story/deepseek-took-53-of-openrouter-s-27t-daily-tokens-a-post-reports/
