GPT-5.4 ProOpenAI · since Apr '26
$30 input$180 output
LLM API Pricing Trends
Track how LLM API costs have evolved since GPT-4's launch in March 2023. Prices are per million tokens based on public API pricing from each provider.
Data from public pricing pages and announcements. Last updated: 2026-07-02.
New: this data now powers the BenchLM Token Price Index — one monthly number for where AI token prices stand, published with a full methodology.
Biggest Price Drop
GPT-4 Output: $60 → $10
83% drop via GPT-4o replacement
Cheapest Frontier
DeepSeek V3
$1.1/1M output · DeepSeek
Price Index
100 → 5.5
94.5% drop since March 2023
LLM Price Index (March 2023 = 100)
Average output price across frontier models, normalized to GPT-4 launch pricing. Lower means cheaper.
Output Price History by Model
Output price per million tokens over time. Toggle models to compare.
OpenAIAnthropicGoogleDeepSeekMistralxAI
Current LLM API prices
Per million tokens, highest output price first
Model / providerInput / output
Claude 3 OpusAnthropic · since Mar '24
$15 input$75 output
GPT-4OpenAI · since Mar '23
$30 input$60 output
o1OpenAI · since Sep '24
$15 input$60 output
GPT-4 TurboOpenAI · since Nov '23
$10 input$30 output
Claude Opus 4.6Anthropic · since Feb '26
$5 input$25 output
GPT-5.4OpenAI · since Apr '26
$2.5 input$15 output
Claude 3 SonnetAnthropic · since Mar '24
$3 input$15 output
Claude 3.5 SonnetAnthropic · since Jun '24
$3 input$15 output
Claude Sonnet 4.5Anthropic · since Sep '25
$3 input$15 output
Claude Sonnet 4.6Anthropic · since Feb '26
$3 input$15 output
Grok 4.1xAI · since Feb '26
$3 input$15 output
GPT-5.2OpenAI · since Feb '26
$1.75 input$14 output
Gemini 3.1 ProGoogle · since Apr '26
$2 input$12 output
GPT-4oOpenAI · since May '24
$2.5 input$10 output
GPT-5.1OpenAI · since Sep '25
$1.25 input$10 output
Gemini 2.5 ProGoogle · since Jun '25
$1.25 input$10 output
o3OpenAI · since Apr '25
$2 input$8 output
GPT-4.1OpenAI · since Apr '25
$2 input$8 output
Claude Haiku 4.5Anthropic · since Sep '25
$1 input$5 output
Gemini 1.5 ProGoogle · since May '24
$1.25 input$5 output
GPT-5.4 miniOpenAI · since Apr '26
$0.75 input$4.5 output
o3-miniOpenAI · since Apr '25
$1.1 input$4.4 output
Gemini 3 FlashGoogle · since Dec '25
$0.5 input$3 output
Gemini 2.5 FlashGoogle · since Dec '24
$0.3 input$2.5 output
DeepSeek R1DeepSeek · since Jan '25
$0.55 input$2.19 output
GPT-4.1 miniOpenAI · since Apr '25
$0.4 input$1.6 output
Mistral Large 3Mistral · since Feb '25
$0.5 input$1.5 output
Claude 3 HaikuAnthropic · since Mar '24
$0.25 input$1.25 output
DeepSeek V3DeepSeek · since Jan '25
$0.27 input$1.1 output
GPT-4o miniOpenAI · since Jul '24
$0.15 input$0.6 output
Grok 3 MinixAI · since Apr '25
$0.3 input$0.5 output
GPT-4.1 nanoOpenAI · since Apr '25
$0.1 input$0.4 output
Gemini 1.5 FlashGoogle · since May '24
$0.075 input$0.3 output
Prices per million tokens from official API pricing. Does not include batch, cached, or volume discounts.
Frequently Asked Questions
How much have LLM API prices dropped?
Since March 2023, the average output price for frontier LLMs has dropped approximately 94.5%. Our price index fell from 100 to 5.5, driven by competition and efficiency improvements across all major providers.
Which LLM provider is cheapest right now?
DeepSeek currently offers the lowest per-token pricing among frontier-capable models, with DeepSeek V3 at $1.10/1M output tokens. Google and OpenAI also offer competitive pricing through Gemini 2.5 Flash and GPT-4o mini respectively.
Why do reasoning models cost more?
Reasoning models like o1, o3, and DeepSeek R1 generate internal chain-of-thought tokens before producing the final answer. These thinking tokens consume compute but are billed as output tokens, making the effective cost per useful output token significantly higher.
Will LLM prices keep dropping?
The trend over the past three years has been consistent price decreases driven by model distillation, hardware improvements, and provider competition. While frontier reasoning models may maintain premium pricing, standard inference prices are expected to continue declining.
Choose a model with this week’s evidence
Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.
One email each week. Unsubscribe anytime.