Spend · 2026-10-14
Follow the money.
LLM token volume is one race; dollars are another. Cheap AI models win volume, premium models keep the revenue. The Token Price Index v2 is the average price of a token routed through the gateway, weighted by real per-model volume (base = 100 on 2025-01-01).
Token Price Index v2
244.2
+17.4 pts · 30d — the average token is more expensive.
Median $ / M paid
$0.269
Usage-weighted blended (30/70 prompt/completion). Absolute-dollar companion to TPI.
What TPI means
If the average token cost $1.00 on 2025-01-01, it costs $2.44 today — weighted by what people actually use.
How this is computed →Token volume vs. average token price
Daily token traffic, estimated daily spend, and the share-weighted Token Price Index. Compare how much is being used with what the average token costs.
Priced models cover 97.8% of tracked tokens on 2026-10-14. Unpriced models are excluded, never counted as $0.
Volume vs price
Average token price fell 60.2% over the last 90 days (since 2026-07-16); estimated daily spend rose 53.3% over the last 30 days. Source: OpenRouter (openrouter.ai/rankings), as of 2026-10-14. tidelines.ai/spend
Tracked tokens / day
25T
Est. spend / day
$33.1M
Latest single day at OpenRouter catalog prices — ignores discounts, cached tokens, and free traffic.
Token share vs dollar share, by lab
Left bar is how much traffic a lab captures. Right bar is our estimate of how many of the dollars come home.
Total tracked tokens grew 53559.5% over the last 652 days of archive to 26.37T/day — these share swings sit on top of a expanding pie.
How to read: Each row is one lab. The left bar shows its share of tokens; the right bar shows its share of estimated dollars. Wider gaps between the two mean cheap volume or premium revenue.
- 7.2%Anthropic+29.336.5%
- 24.9%DeepSeek-19.05.8%
- 15.2%StepFun+0.315.4%
- 11.0%Z-AI-3.37.7%
- 8.2%OpenAI+2.410.6%
- 1.6%Moonshot+6.37.9%
- 7.6%Xiaomi-6.51.1%
- 6.4%Tencent-1.94.4%
- 6.1%Other-5.80.3%
- 5.8%Nvidia-1.93.8%
Window: latest single day (2026-10-14). Dollar share = a lab's dollars ÷ all priced dollars that day; dollars = tokens × OpenRouter catalog price (30% prompt / 70% completion). Unpriced models are excluded from both sides. Estimate, not observed revenue.
Value map: price vs momentum
Every tracked model plotted by price (x, log) and 30-day change in share (y). Bubble size is current share.
How to read: X-axis is list price on a log scale; Y-axis is 30-day change in share; bubble size is current share. The upper-left quadrant — cheap and rising — is the winners' club.
As of 2026-10-14