How much prompt caching (cache reads ~90% off input) and the Batch API (50% off) cut your AI bill on OpenAI, Claude and Gemini, with a worked example from live prices.
| Vendor | Cache read | Cache write | Batch API |
|---|---|---|---|
| OpenAI | 0.1× input (0.05× on GPT-6.1 Sol/Terra) | 1.25× input (30-min), 2.5× (12-hour) | 50% off, within 24 h |
| Anthropic (Claude) | 0.1× input (0.05× Opus/Sonnet 5.5) | 1.25× (5-min), 2× (1-hour) | 50% off; stacks with caching |
| Google (Gemini) | Discounted, automatic (implicit caching) | Explicit caches add hourly storage | 50% off, 24 h target |
10,000 monthly users × 30 messages, 600 words in / 150 words out per message, assuming 70% of input tokens are a repeated prompt prefix served from cache (cache-write premium ignored). Cache reads priced at a conservative 0.1× input:
| Model | List price | With caching | Batch only | Caching + batch |
|---|---|---|---|---|
| OpenAI: GPT-6.1 Sol | $1,077 | $775.66 | $538.65 | $387.83 |
| Anthropic: Claude Sonnet 5.5 | $1,077 | $775.66 | $538.65 | $387.83 |
| Google: Gemini 4 Argon | $1,077 | $775.66 | $538.65 | $387.83 |
| Anthropic: Claude Haiku 5.5 | $53.87 | $38.78 | $26.93 | $19.39 |
When it applies: caching needs an identical prompt start (system prompt, documents, tools). Batch only suits jobs that can wait hours: backfills, evals, nightly summaries, not live chat.
Sources (checked 2026-10-31): OpenAI prompt caching · Anthropic prompt caching · Gemini Batch API · Prices: OpenRouter
Last updated 2026-10-30 from live API prices.
Cached input tokens are typically billed at 10% of the normal input price (even 5% on some flagship models), so a prompt that is mostly a repeated prefix can cut input cost by up to ~90%.
50% on input and output at OpenAI, Anthropic and Google, in exchange for results within up to 24 hours.
Anthropic says the discounts stack; OpenAI's batch price table halves cached rates too. Cache hits inside batches are best-effort.
One short email a week on which AI API prices changed and what it means for your bill. Subscribers get the free AI API Cost Cheat Sheet (1-page PDF).
Free · Unsubscribe anytimeTry the interactive calculators → · All guides
Every week: which AI API prices changed, and what it means for your app's bill. Free, plus the AI API Cost Cheat Sheet.