LLM API Price History 2024–2026: How AI Costs Dropped 90%
A historical tracker of list prices for the major LLM APIs. Every figure is taken from the provider's own pricing page at the listed date — no estimates, no scraping of aggregator sites. Free to cite; see methodology.
By the LLMCost data team · Published 2026-07-01 · Last reviewed 2026-07-21
Headline insight
The cost of calling a frontier LLM has fallen roughly 90% since 2023. GPT-4 input pricing alone dropped 92% ($30 → $2.50/1M tokens). For a workload that cost $1,000/month in 2023, the equivalent frontier model now costs under $100/month at list price — before any volume discount.
OpenAI — GPT-4 family
−92% input| Model | Date | Input / 1M | Output / 1M |
|---|---|---|---|
| GPT-4 | Mar 2023 | $30.00 | $60.00 |
| GPT-4 Turbo | Nov 2023 | $10.00 | $30.00 |
| GPT-4o | May 2024 | $5.00 | $15.00 |
| GPT-4o (2026) | Jul 2026 | $2.50 | $10.00 |
Flagship input pricing has fallen 92% in three years — from $30 to $2.50 per million tokens.
Anthropic — Claude family
−63% input| Model | Date | Input / 1M | Output / 1M |
|---|---|---|---|
| Claude 2 | Jul 2023 | $8.00 | $24.00 |
| Claude 3 Opus | Mar 2024 | $15.00 | $75.00 |
| Claude 3.5 Sonnet | Jun 2024 | $3.00 | $15.00 |
| Claude Sonnet 4 | 2025 | $3.00 | $15.00 |
Claude 3 Opus was a price outlier; Anthropic has since routed flagship capability through the far cheaper Sonnet line.
Google — Gemini family
+150% input| Model | Date | Input / 1M | Output / 1M |
|---|---|---|---|
| Gemini Pro 1.0 | Dec 2023 | $0.50 | $1.50 |
| Gemini 1.5 Pro | Feb 2024 | $1.25 | $5.00 |
| Gemini 2.5 Pro | 2025 | $1.25 | $10.00 |
Google entered below cost to gain share, then raised Pro-tier output pricing as context windows grew.
DeepSeek — DeepSeek family
+93% input| Model | Date | Input / 1M | Output / 1M |
|---|---|---|---|
| DeepSeek V2 | May 2024 | $0.14 | $0.28 |
| DeepSeek V3 | Dec 2024 | $0.27 | $1.10 |
| DeepSeek V3 (2026) | Jul 2026 | $0.27 | $1.10 |
The cheapest frontier-class API on the market — roughly 1/10th the input cost of GPT-4o.
Visual: GPT-4 input price collapse
Bar height is proportional to list input price (USD / 1M tokens). The story is one direction: down.
The Claude family tells the same story at smaller scale: a 80% input drop from Claude 3 Opus to Claude Sonnet 4.
What This Means for Businesses
- 1.Re-evaluate build-vs-buy every quarter. A feature that was too expensive to ship on GPT-4 in 2023 may now be comfortably profitable on GPT-4o or DeepSeek V3. Re-run your unit economics.
- 2.Model the workload, not the model. Because output tokens cost 2-5× input, a small model that produces fewer output tokens can beat a cheaper model on total bill. Enter your real volumes in the calculator to compare.
- 3.Lock in with caution.List prices are still falling. Long committed-use discounts only win if they beat the trajectory — a 30% discount locked for 12 months can be undercut by next quarter's list cut.
- 4.Cache aggressively. Prompt caching and context-cached rates can cut effective input cost a further 50-90% on repeat-heavy workloads — the cheapest token is the one you never send.
Methodology & Sources
Every price in this tracker is a public list rate sourced directly from the provider's official pricing page at the listed revision date. No figures are scraped from pricing aggregators or third-party blogs.
Primary sources
- OpenAI — openai.com/api/pricing/ (archived snapshots at each model launch: GPT-4 Mar 2023, GPT-4 Turbo Nov 2023, GPT-4o May 2024).
- Anthropic — anthropic.com/pricing (Claude 2 Jul 2023, Claude 3 Opus Mar 2024, Claude 3.5 / Sonnet 4 subsequent revisions).
- Google DeepMind — ai.google.dev/pricing (Gemini Pro 1.0, 1.5 Pro, 2.5 Pro).
- DeepSeek — api-docs.deepseek.com/quick_start/pricing (V2 May 2024, V3 Dec 2024).
Conventions
- All prices are USD per 1,000,000 tokens, list rate, at the model's launch or listed revision date.
- Prices exclude volume tier discounts, committed-use discounts, and cached-prompt / context-cached rates.
- Where a provider changed price mid-generation, the most recent publicly listed rate for that generation is used.
- Drop percentages are computed from list input price only; output drops are noted separately where material.
Review cycle: monthly. Next scheduled refresh: August 2026. When citing, please link to this page and note the review date.
Frequently Asked Questions
How much has the GPT-4 API price dropped?
GPT-4 launched in March 2023 at $30 per million input tokens and $60 per million output tokens. By July 2026 the successor GPT-4o lists at $2.50 input and $10 output — a 92% drop in input pricing and an 83% drop in output pricing over three years.
Is AI getting cheaper every year?
Yes. Across the four major providers we track, flagship-class input pricing has fallen roughly 60-90% per generation since 2023. Two forces drive this: inference hardware efficiency (per-token compute cost halves roughly every 12-18 months) and aggressive below-cost pricing by newer entrants competing for developer share.
What is the cheapest LLM API in 2026?
DeepSeek V3 remains the cheapest frontier-class model in 2026 at $0.27 per million input tokens. For workloads that do not need frontier reasoning, Gemini 2.0 Flash and Mistral Small are comparable or cheaper. Use our calculator above to find the cheapest model for your specific token mix.
Why did Claude 3 Opus cost more than Claude 2?
Claude 3 Opus (March 2024) was positioned as Anthropic's top reasoning tier and priced at $15/$75 — higher than Claude 2's $8/$24. Anthropic subsequently moved flagship capability into the Sonnet line at $3/$15, so the practical cost of top-tier Claude access dropped ~80% despite the brief Opus spike.
How accurate are these historical prices?
Every figure is sourced from the provider's official pricing page at the listed date (archived snapshots linked in the methodology). Prices are public list rates per 1M tokens in USD and exclude volume discounts, committed-use deals and cached-prompt rates, all of which lower the effective price further.
Cite this data
Free to reference in articles, reports and analyses. Suggested citation:
LLMCost (2026). LLM API Price History 2024–2026. https://llmcost-8k0.pages.dev/price-tracker/ Accessed 2026-07-21.