The price of the same job

Data Desk · Part 05AI Labs

The price of the same job

One coding task, three models. On a linear scale the cheapest one all but disappears. Switch to a log scale, then dial up the volume to see a buyer’s monthly bill.

As of Aug 9, 2026 · with each list-price change · published Aug 9, 2026 · read Sep 11, 2026 · scenario

What the chart says. Same task, a 64-fold price gap. A general-purpose model cannot charge frontier prices for work a router can send elsewhere. The moat has to be the work only you can do.

Source: Wyndo (cost of the same coding task, August 9, 2026), as reported in Part 05. Monthly bills are DeepStack arithmetic.

Show the data table
ModelCost per task1,000 tasks10,000 tasks100,000 tasks1,000,000 tasks10,000,000 tasks
Claude Opus 5$0.32$320.00$3,200$32,000$320,000$3.20M
Kimi$0.24$240.00$2,400$24,000$240,000$2.40M
DeepSeek$0.005$5.00$50.00$500.00$5,000$50,000
Read Part 05: OpenAI and Anthropic Are Not the Same Business
1,000 tasks a month10,000,000 tasks a month
Step 10x the volume
  1. Claude Opus 5$0.32 per task$320,000a month
  2. Kimi$0.24 per task$240,000a month
  3. DeepSeek$0.005 per task$5,000a month

Linear scale: bar length is proportional to price.

At 1,000,000 tasks a month: Claude Opus 5 $320,000, Kimi $240,000, DeepSeek $5,000. The most expensive model costs 64 times the cheapest.