openrouter top 5 by token volume. every model gaining share on openrouter this week is either chinese or free

1. glm 5.3 flash – 19t ↑67%
2. deepseek v4.1 flash – 18.9t ↑60%
3. hy4 preview – 12.3t ↑6%
4. gpt-5.6 luna – 8.7t ↓45%
5. deepseek v4 flash – 8.3t ↓22%

the two at the top are also the two cheapest on the board – $0.045 and $0.04 per million input tokens. hy4 preview costs 20x more and grew 6%.

nvidia's nemotron 3 ultra is the highest-placed non-chinese model still gaining – at #6, up 37%, and free.

gpt-5.6 luna is the only paid us model in the top 5, and it's the steepest drop on the board – down 45% as gpt-6 luna landed at half the input price.

builders aren't renting the frontier tier by the token anymore – they rent the cheapest thing that finishes the job.