How are token costs calculated?

Multiply input tokens by the input rate and output tokens by the output rate, add the two amounts and divide by 1,000,000. A standard request example uses 1,000 input and 500 output tokens. Monthly budgets multiply the result by request volume; taxes, hidden reasoning tokens, tools and tier surcharges can add costs.

Which recorded model has the lowest request cost?

lfm-2.2-6b by Liquid costs $0.00002 for the standard request, with $0.01 input and $0.02 output per million tokens. This price ranking does not establish quality equivalence; verify task success and context requirements.

Can cached and batch discounts be combined?

Only recorded rates are used. Cached input can have cache-write or storage charges, while batch rates apply only to eligible asynchronous jobs. We do not automatically stack the discounts or treat an absent rate as zero. Check the linked provider source before budgeting.

How current is the public dataset?

The live pricing table synchronizes stored model records daily; public file snapshots update when the site is rebuilt. Synchronization is not independent provider verification. It contains 1141 priced models across 131 providers, last updated 2026-10-31. Directory entries are counted separately. A synchronization date is not proof of independent verification.

Analysis · 2026-10-31 · 7 min read

AI Spend per Token Fell 21% While Google Committed $924B: What It Means for Your Bill

Ramp data shows business AI token volume up 37.1% in four weeks while spend rose only 8.9%, about 21% less per token. Meanwhile Alphabet lists $924.4B in purchase commitments. Here is the TokenCost read for buyers.

Tokens Up 37%, Spend Up 9%

Businesses paying through Ramp used 37.1% more AI tokens in the week ending October 23, 2026 than in the same week a month earlier, while their token spending rose only 8.9%, according to Ramp's AI Index. As FourWeekMBA calculated, that works out to about 21% less spend per token in four weeks.

The simple math: 1.089 ÷ 1.371 ≈ 0.794. Each token cost roughly 79% of what it did a month earlier.

Blended Prices: OpenAI Fell Fastest

Ramp's table gives a blended price per million tokens across what its customers actually bought:

*Prior-week values for the individual providers are derived from Ramp's stated percentage changes.*

Why is Ramp's blended drop (26.8%) bigger than the 21% spend-per-token figure? Ramp computes its totals and its blended price separately, so the two don't line up exactly. Either way, the direction is clear.

Important: these blended prices are far below list prices. They mix cached input, fresh input, output and cheap small models. A blended $0.56/M doesn't mean flagship models cost $0.56/M. It means buyers are routing more traffic to cheap tiers and cached reads. Compare with list prices on our pricing table.

Anthropic Still Takes the Bigger Share

In the latest week, Anthropic accounted for 60.5% of Ramp customers' token spend and OpenAI for 35.1%. Anthropic's blended price fell less (−15.7%) than OpenAI's (−42.4%), which suggests buyers keep paying a premium for Claude workloads such as coding agents, while OpenAI usage is shifting toward cheaper models and cached traffic.

The Other Side: Alphabet's $924.4 Billion

While buyers pay less per token, the builders are locking in spending for years. Alphabet's Q3 2026 filings show:

$924.4B in material purchase commitments as of September 30, 2026
$213.4B of that due within 12 months, about 5x its Q3 operating cash flow
$802.1B tied to multi-year technical infrastructure, inventory, energy take-or-pay and content contracts
$130.5B more in future lease payments not yet started, mainly data centers and energy
$51.4B of Q3 capex against $42.3B operating cash flow (about 121%), so free cash flow was −$9.1B

The 10-Q says the "significant majority" of these commitments will be fulfilled through 2031.

The TokenCost Read: Two Clocks

Token prices move on a weekly clock. Infrastructure commitments move on a multi-year clock. That mismatch matters for anyone budgeting AI:

1. Prices are falling now, because supply is being built ahead of demand. Hyperscalers are competing to fill capacity they've already committed to. That's good for buyers in the short term.

2. Commitments create pressure to monetize. $213B due in a year has to be paid for. Expect providers to keep cutting per-token prices to grow volume, while pushing premium tiers, priority processing and long-context surcharges where margins are better.

3. Your bill follows volume, not unit price. Volume grew about 4x faster than spend. Cheaper tokens rarely mean a smaller invoice; they usually mean more usage.

What Buyers Should Do

Track effective cost per token, not list price. Divide your monthly spend by total tokens, the same metric Ramp uses. If it isn't falling while market prices are, you're leaving savings on the table.
Re-check model routing monthly. A 42% drop in four weeks means last month's cheapest option may not be this month's. See the three-tier market breakdown.
Max out caching. Much of the blended price drop comes from cached reads. Read Prompt Caching Explained.
Avoid long lock-ins at today's prices. If providers are cutting prices this fast, multi-year committed-spend deals should include price-drop clauses.
Set spend caps. Falling unit prices encourage more usage. Budgets should be set on total dollars, not tokens.

Caveats

Ramp's data only covers businesses that pay through Ramp, not the whole market. It compares two single weeks. Alphabet's commitments include energy and content licenses, not only AI. Neither source links buyer prices to Alphabet's spending; this is our reading of the two together.