このページはまだ日本語で提供されていないため、以下の本文は英語です。英語のページを開く

AI model discounts

Language model vendors publish two ways to pay less: prompt caching and batch processing. This page shows each model's list price beside the price with those discounts, using only published rates.

Settings

Share of input read from the cache: 50%The saving depends on how much of each prompt repeats. A long fixed instruction sent with every request repeats; a new document each time does not.

Currency

Model tier

Prices as of 2026-11-03, per million tokens in USD. These are the discounts vendors publish, not a measure of what it costs them to serve tokens.

Model	Tier	List input	List output	Input at 50% cached	Batch input / output	Last verified
Claude Haiku 5.5Anthropic	Small	$0.1000	$0.5000	$0.0550 (saves 45.0%)	$0.0500 / $0.2500 (saves 50%)	2026-11-03
Claude Sonnet 5.5Anthropic	Mid	$2.00	$10.00	$1.05 (saves 47.5%)	$1.00 / $5.00 (saves 50%)	2026-11-03
Claude Opus 5.5Anthropic	Frontier	$4.00	$20.00	$2.10 (saves 47.5%)	$2.00 / $10.00 (saves 50%)	2026-11-03
GPT-6 LunaOpenAI	Small	$0.1000	$0.5000	$0.0550 (saves 45.0%)	$0.0500 / $0.2500 (saves 50%)	2026-11-03
GPT-6.1 TerraOpenAI	Mid	$1.00	$5.00	$0.5250 (saves 47.5%)	$0.5000 / $2.50 (saves 50%)	2026-11-03
GPT-6.1 SolOpenAI	Frontier	$2.00	$10.00	$1.05 (saves 47.5%)	$1.00 / $5.00 (saves 50%)	2026-11-03
Gemini 3.5 Flash-LiteGoogle	Small	$0.3000	$2.50	$0.1650 (saves 45.0%)	$0.1500 / $1.25 (saves 50%)	2026-11-03
Gemini 3.8 FlashGoogle	Mid	$0.8250	$4.13	$0.4538 (saves 45.0%)	$0.3754 / $1.88 (saves 54.5%)	2026-11-03Check
Gemini 4 ArgonGoogle	Frontier	$2.00	$10.00	$1.05 (saves 47.5%)	$1.00 / $5.00 (saves 50%)	2026-11-03
Llama 3.3 70B InstructMeta	Mid	$0.1000	$0.3200	Not offered	Not offered	2026-11-03
Llama 4 MaverickMeta	Mid	$0.1875	$0.6525	$0.1188 (saves 36.7%)	Not offered	2026-11-03
Qwen3.8 27BAlibaba (Qwen)	Small	$0.1500	$1.88	$0.0938 (saves 37.5%)	Not offered	2026-11-03
DeepSeek V4 FlashDeepSeek	Small	$0.0910	$0.1820	$0.0546 (saves 40%)	Not offered	2026-11-03
DeepSeek V4 ProDeepSeek	Frontier	$1.60	$3.20	$0.8675 (saves 45.8%)	Not offered	2026-11-03
Mistral Small (2603)Mistral	Small	$0.1500	$0.6000	$0.0825 (saves 45.0%)	Not offered	2026-11-03
Mistral Large 4Mistral	Mid	$0.7000	$2.10	$0.3850 (saves 45.0%)	Not offered	2026-11-03

How caching works: the vendor keeps the start of a prompt for a short time, and a repeat of that start is billed at the cache read price instead of the full input price. Some vendors also charge to write to the cache. The saving shown is for the share you choose above, on input only; output is never cached.