gpt 5.6 luna is the first us model in the openrouter top 3 since june. just in 2 weeks after a 80% price cut from outside the top 10

Addy Crezee 14 Aug 2026 1 min read

Soldier planting OpenAI flag on cracking #3 podium block while Chinese flags fly over #1 and #2 in graffiti-covered arena

openrouter top 5 by token volume this week:

1. deepseek v4 flash 0731 – 10.7t
2. tencent hy3 – 10.6t
3. gpt-5.6 luna – 5.2t
4. deepseek v4 flash 0423 – 5.11t
5. xiaomi mimo-v2.5 – 4.3t

the top two both exploded – +74% and +87% in a single week. tencent hy3 is now 100b tokens behind deepseek v4 flash. if the trend holds, tencent takes #1 next week.

two deepseek flash versions in the top 5 – builders haven't fully migrated to the new release. all 5 under $0.15/m input. luna at $0.10 – cheaper than most chinese models on this list.

anthropic held a top-10 spot for months – now it's gone. deepseek holds 3 spots in the top 10. openai bought one with a price cut.

the price war worked and the volume game belongs to cheap inference.

OpenRouter top 10 models by token usage table showing DeepSeek V4 Flash at 10.7T tokens, Tencent HY3 at 10.6T, and GPT-5.6 Luna at 5.2T

glm 5.3 vs kimi k3 – four hogwarts locations in three.js

the setup: streamed straight off @openrouter, no agent loop, no tools, no filesystem – one prompt in, one html file out. the system prompt is a three.js skill, 32,014

21 Aug 2026 1 min read

Illustrated man in jacket with Chinese flag patches holds cash stack, hands $6 receipt to server rack in neon-lit data center

the performance gap between us and chinese models is at a record-low 6% per bloomberg

@rachelmetz, ai reporter at @technology, broke down the cost math: the industry is quietly moving from cost per token to cost per task. not how cheap the model is per

21 Aug 2026 1 min read