OpenRouter's Usage and Spend Charts Have Split: Used Models Aren't the Paid Ones
Spiritual_Skirt_9312 · reddit · 2026-10-08
Tracking OpenRouter rankings for three weeks, the author finds token volume and spending have come apart.
Usage side: Chinese models dominate — DeepSeek V4.1 Flash leads at 33.6T tokens (+48% WoW), with identifiable Chinese labs accounting for 62% of top-10 volume; the top US model sits at #6, and the China/US volume ratio holds steady at 3.4–3.5x. The #2 entry, Space Bunny Alpha (28T tokens), has no credible attribution.
Spend side (Classification category, 11% of platform spend): eight of the top nine are American (GPT-5.6 Sol, GPT-6 Astra, Gemini 3.1 Pro lead), with Kimi K3 the sole Chinese entry; US models churn fast — Opus 5 lost 4.9 points in a week while Opus 5.5 gained 4.3.
Thesis: usage and revenue are decoupling into two layers — cheap open weights absorb traffic while money flows up to compute, cloud, and enterprise contracts. The author concedes the counterargument: US models are climbing fast (Opus 5.5 +74% WoW), which could mean this is arbitrage being corrected rather than a structural shift.
More from Venture
- Proximal reportedly hitting $55M in monthly revenue amid preemptive funding — nrehiew_ · 2026-10-08
- Building Is Cheap, Distribution Is the Only Moat Left in AI, Argues victor_explore — victor_explore · 2026-10-08
- AI series episode-3 intro produced for just $200, creator shows the making — Pressmediaphp · 2026-10-08
- Ex-ByteDance intern Keyu Tian raises ~$30M at $200M valuation for world models lab — zephyr_z9 · 2026-10-08
- Firmus AI datacenter startup slashes $44B valuation as ASX IPO wobbles — nordicinst · 2026-10-08
- Lessons from Gamma's 100M users: consumers won't pay for AI subscriptions — omooretweets · 2026-10-08