Today
Gpt 6 Astra
204B tokens routed
30.27%
OpenRouter data
Model rankings
Which models are actually moving tokens through the market. Rankings are based on OpenRouter usage data and should be read as market signal, not absolute internet-wide usage.
Week
Gpt 6 Astra
932B tokens routed
28.11%
Month
Gpt 6 Astra
1.28T tokens routed
27.52%
Trending
Llama 4 Free
732B tokens routed
Today
Today leaderboard
#1
Gpt 6 Astraopenai/gpt-6-astra
204B · 30.27%#2
Llama 4 Freemeta/llama-4-free
181B · 26.85%#3
Gemini 3 Progoogle/gemini-3-pro
151B · 22.4%#4
Claude Sonnet 5anthropic/claude-sonnet-5
82B · 12.17%#5
Deepseek V4deepseek/deepseek-v4
56B · 8.31%Week
Week leaderboard
#1
Gpt 6 Astraopenai/gpt-6-astra
932B · 28.11%#2
Gemini 3 Progoogle/gemini-3-pro
785B · 23.67%#3
Llama 4 Freemeta/llama-4-free
732B · 22.07%#4
Claude Sonnet 5anthropic/claude-sonnet-5
487B · 14.69%#5
Deepseek V4deepseek/deepseek-v4
380B · 11.46%Month
Month leaderboard
#1
Gpt 6 Astraopenai/gpt-6-astra
1.28T · 27.52%#2
Gemini 3 Progoogle/gemini-3-pro
1.04T · 22.25%#3
Llama 4 Freemeta/llama-4-free
868B · 18.6%#4
Claude Sonnet 5anthropic/claude-sonnet-5
755B · 16.18%#5
Deepseek V4deepseek/deepseek-v4
721B · 15.45%Trending
Fastest movers
#1
Llama 4 Freemeta/llama-4-free
+438.2%#2
Gemini 3 Progoogle/gemini-3-pro
+210.3%#3
Gpt 6 Astraopenai/gpt-6-astra
+164.8%#4
Claude Sonnet 5anthropic/claude-sonnet-5
+81.7%#5
Deepseek V4deepseek/deepseek-v4
+11.4%Source: OpenRouter rankings data. The data is republished with attribution and cached so the page keeps working through short upstream outages.
Which models developers actually route traffic to, measured by token volume rather than opinion or benchmark score.
About this data
What is being measured here?
Tokens processed through OpenRouter over the period shown. It is a usage signal from real applications, which is a different thing from either a benchmark score or a popularity vote.
Does this represent the whole market?
No. It reflects traffic through one router. Developers calling OpenAI or Anthropic directly are not counted, so treat it as a strong directional signal about what builders choose rather than a complete market share figure.
Why does a cheap model rank so highly?
Volume follows cost. High-throughput work — classification, extraction, summarising at scale — gravitates to the cheapest model that is good enough, so token share rewards cheap models more than capable ones.