A monthly benchmark, recomputed every day from published model prices. If your AI bill was set more than a month ago, the number on it is out of date — in both directions.
A 40-person company running 120M input tokens and 18M output tokens a month bills anywhere from $3.65 (cheapest capable model) to $840 (90th percentile) for the same volume. That is a 7.7x gap on an input nobody negotiates, across 452 models listed publicly today. The median bill for that volume is $110 a month — so a company sitting at the top of the range is overpaying roughly $730 a month — $8,766 a year — against the middle of the market.
| Company | Monthly volume | Cheapest workhorse | Median of market | Premium (90th pct) | Annualised at premium |
|---|---|---|---|---|---|
| 10-person shop a support bot, some drafting, a little code help |
25M in / 3M out | $0.71 | $21.32 | $160 | $1,920 yr |
| 40-person company everyone has a seat, one product feature uses a model |
120M in / 18M out | $3.65 | $110 | $840 | $10,080 yr |
| 150-person company seats plus two production features and a document pipeline |
500M in / 90M out | $16.17 | $483 | $3,800 | $45,600 yr |
Method: every one of the 452 models in OpenRouter's public catalogue with a context window of 100k tokens or more (388 of them), priced at the published rate per million input and output tokens, applied to the stated monthly volume. The ceiling is far higher than the 90th percentile — the most expensive workhorse model today would bill a 40-person company $6,840 a month for the same volume. Prices change daily; this page is rebuilt each morning from the same log the briefs come from.
Cheap is not the same as equivalent. The cheapest models in this table are small, fast and fine for classification, extraction, drafting and support replies; they are not what you write a long-context agent on. The point is not "use the cheapest model" — it is that most companies are paying premium rates for the routine share of their traffic, and the routine share is usually most of it. Splitting traffic by task is where the money is.
| Model | Field | Was /M | Now /M | Change |
|---|---|---|---|---|
| Qwen: Qwen3.8 27B | input | $0.1500 | $0.0400 | -73.3% |
| Z.ai: GLM 5.3 | output | $3.6000 | $1.4000 | -61.1% |
| Particle AI: Particle-1 | input | $0.1000 | $0.0500 | -50.0% |
| Particle AI: Particle-1 | output | $1.0000 | $0.5000 | -50.0% |
| DeepSeek: DeepSeek V4 Flash Latest | output | $0.3000 | $0.1729 | -42.4% |
| DeepSeek: DeepSeek Flash Latest | input | $0.0220 | $0.0141 | -35.9% |
AI Price Watch · re-bid · my own audit · local transcription · clinic · feed
Source: OpenRouter public model catalogue, logged daily. Prices in USD per million tokens. This is a factual index of published list prices, not advice, and not a quote for any vendor's contract. Updated 2026-10-31.