Verified November 2026

Home › Best › Lowest output price

Lowest output token price

Output tokens are billed at a premium and are usually the easiest count to reduce. Ranked by output price per million tokens. Cached input and batch discounts do not reduce output cost.

Top 15 by output price

#ModelProviderOutput $/MInput $/MContext
1GPT-6 LunaOpenAI$0.50$0.101,050,000
2Claude Haiku 5.5Anthropic$0.50$0.10200,000
3Mistral Small 4Mistral AI$0.60$0.15256,000
4GPT-5.6 LunaOpenAI$1.20$0.201,050,000
5DeepSeek V4.1 FlashDeepSeek$1.20$0.301,100,000
6GPT-5.4 nanoOpenAI$1.25$0.20400,000
7Gemini 3.1 Flash-LiteGoogle$1.50$0.251,000,000
8Mistral Large 4Mistral AI$2.10$0.701,000,000
9Gemini 3.5 Flash-LiteGoogle$2.50$0.301,000,000
10Grok 4.3xAI$2.50$1.251,000,000
11Gemini 3.8 FlashGoogle$3.75$0.751,000,000
12Gemini 3.6 FlashGoogle$3.75$0.751,000,000
13DeepSeek V4 ProDeepSeek$3.96$1.321,000,000
14GPT-5.4 miniOpenAI$4.50$0.75400,000
15GPT-6.1 TerraOpenAI$5$11,050,000

Frequently asked

Why is output more expensive than input?

Generation is compute-heavy and sequential, while input can be processed in parallel. Every provider in this dataset charges more for output than input, typically 3-5x.

How do I reduce output cost?

Cap max_tokens, use a cheaper model for drafting and a stronger one only for final answers, and avoid asking for reasoning tokens where the answer is short. Output is the one lever you control directly.

Do cached or batch rates apply to output?

Batch discounts apply to output (roughly 50% off). Prompt caching never discounts output, only input.

Browse all 45 models · Compare two models