Prompt caching.
Anthropic cache reads cost 0.1x base input (0.05x on Opus/Sonnet 5.5; 0.025x on Fable 5.1). A 5-min cache write costs 1.25x, so it pays off after one read. OpenAI cached input e.g. gpt-6.1-sol $0.10 vs $2.00.
Spend less, build more
Cost-saving tips
Sourced; checked 14 Oct 2026
Batch APIs, 50% off.
Anthropic Batch API is 50% off input and output. OpenAI Batch pricing is half of standard (e.g. gpt-6.1-sol $1/$5). Gemini Batch is a "50% cost reduction".
Stack caching and batch.
Anthropic says the two discounts can be combined.
Right-size the model.
Anthropic suggests Haiku for simple tasks, Sonnet for most production work and Opus for complex reasoning. Haiku 5.5 is $0.10/$0.50 vs Opus 5.5 at $4/$20.
Use Flex instead of Fast on OpenAI.
Flex costs half of standard; Fast costs 2x (gpt-6.1-sol: $1 vs $4 input).
Avoid long-context surcharges.
OpenAI charges more above 272K input tokens (gpt-6.1-sol $2 → $4 input); Gemini 2.5 Pro charges more above 200K.
Watch for scheduled price rises.
Gemini 4 Argon input doubles from $2 to $4 on 1 Dec 2026, and Gemini 3.8 Flash doubles on 1 Jan 2027.
Account for tokenizer changes.
Claude 4.7+ uses a tokenizer that produces about 30% more tokens for the same text.
Skip US-only routing unless you need it.
It's 1.1x on Claude 4.6+, and OpenAI adds a 10% regional-processing uplift.
Cap fetched content.
Use `max_content_tokens` on Claude web fetch to limit token spend.
Get the full cost playbook
$19, coming soon