vibecoding.tech_
vibecoding.tech_
🗺️AI mapGuide
🗺️AI mapGuide
NewsAI agent rankingComparisonsWhat vibecoding is
vibecoding.tech
© 2026
HomeMapGuideNews📬Newsletter

./open · api-costs

What the agents building this site cost

Agents have been building vibecoding.tech since July 1, 2026, and every message they send sits in the transcripts. The counter multiplies tokens by public prices and puts two totals side by side: what it would have cost via API and what was actually paid through subscriptions.

1 Jul 2026 - 27 Sep 2026 · 89 days · one counter from transcripts
51,1B
tokens burned
$52 366
would cost via API
$1 494
paid via subscriptions
×35
subscriptions cheaper than API
our price $29 per 1B tokens, via API $1 024
AgentModelSession
Claude Code$48,2K
Codex$4 164
Fable 5$28,9K
Fable 5.1$7 896
Opus 5$6 510
Opus 5.5$3 643
Opus 4.8$1 284
GPT-5.6 Sol$2 541
GPT-5.5$1 039
other models$597
Main session$42,3K
Subagents$10,1K
  • Fable 5$28 85555,1%
    17,9B tokens · $28 855 via API · $804 in subscriptions
  • Fable 5.1$7 89615,1%
    9,1B tokens · $7 896 via API · $180 in subscriptions
  • Opus 5$6 51012,4%
    7,7B tokens · $6 510 via API · $125 in subscriptions
  • Opus 5.5$3 6437,0%
    10,0B tokens · $3 643 via API · $40 in subscriptions
  • GPT-5.6 Sol$2 5414,9%
    3,3B tokens · $2 541 via API · $144 in subscriptions
  • Opus 4.8$1 2842,5%
    1,1B tokens · $1 284 via API · $26 in subscriptions
  • GPT-5.5$1 0392,0%
    1,3B tokens · $1 039 via API · $131 in subscriptions
  • GPT-6 Astra$3600,7%
    466M tokens · $360 via API · $34 in subscriptions
  • Codex, model not recorded$1150,2%
    155M tokens · $115 via API · $2 in subscriptions
  • and more 4
by daypeak $2 593
1 Jul 202627 Sep 2026
  • Every model message is counted once, even if a continued session copied the history into a new file.
  • Subagents are included, they have their own column in the stream.
  • Codex is counted by spend events, not by the file's total: a session fork carries the parent's total.
  • Prices from Anthropic's public price lists and the LiteLLM registry for OpenAI; where there is no price, an estimate based on GPT-5.6 Sol is used with a note; cache writes ×1.25 and ×2, reads at the model's cache price.
  • The subscription is spread across days and split between the day's models by their share of API dollars within their own vendor; the counter knows nothing about discounts or invoices and shows only this site.

Breakdown: why the subscription came out 35 times cheaper than API (snapshot Sep 27, 2026)The whole build live: machine, cost, selection