Claude Price History: API and Plan Prices Since 2023
Every Claude API list price since March 2024, plan price and limit changes, a CSV download, and what the newer tokenizer means for real costs.
FSFlorian SchröderNov 1, 202610 min read Human reviewed
✓ ReviewedLast updated November 1, 2026 by Florian Schröder
Claude’s list prices have fallen, but less than the headline numbers suggest. Opus input went from $15 to $4 per million tokens between 2024 and 2026, a 73% drop, while the most capable public tier today, Fable, costs $10. Haiku fell to $0.10 for prompts up to 100K tokens. Since Opus 4.7, a newer tokenizer also produces about 30% more tokens for the same text. Claude Pro costs $20 per month, the same as at its 2023 launch.
Part of our Claude AI guide. Data last checked on November 1, 2026. Every API row links to a source, and plan changes before 2026 come from Anthropic’s announcements, press coverage and archived copies of its pricing page. For today’s full price list, including caching and long-context rates, see our Claude pricing guide.
Opus got much cheaper, in two steps. It cost $15 / $75 per million tokens from March 2024 until Opus 4.5 cut it to $5 / $25 in November 2025. Opus 5.5 brought it to $4 / $20 in September 2026. That is 73% off the list price. Opus is no longer the top tier, though: Fable launched above it at $10 / $50 in June 2026, so from the 2024 flagship to today’s the input price fell 33%.
Sonnet barely moved for over two years. It stayed at $3 / $15 from March 2024 until Sonnet 5 launched at $2 / $10 in June 2026, across seven model releases at the same price. That $2 / $10 began as an introductory price, with a rise to $3 / $15 planned before Anthropic made it standard.
Haiku went up before it came down. Claude 3.5 Haiku launched at four times the price of Claude 3 Haiku, and Haiku 4.5 reached $1 / $5. Haiku 5.5 then cut that by 90% to $0.10 / $0.50 for prompts up to 100K tokens. Above that, it costs $0.50 / $2.50, a 50% cut.
Output has cost five times input at base list price. That ratio holds for all 24 models in this dataset. Long-context rates, where they applied, used a different ratio.
The tokenizer changes the comparison. On a rough like-for-like basis, Opus 5.5 costs about 4% more than Opus 4.6, not 20% less. Against Opus 5, which uses the same tokenizer, the 20% cut is real. Our calculation is below.
Claude API input price over time
The chart follows the newest public model in each family. Hover over a point to see its date, or use the table below.
List prices in US dollars per million input tokens. The logarithmic scale keeps $0.10 and $15 on one chart, so equal distances mean equal percentage changes. Mythos models are left out because their access is limited. Mythos 5 and 5.1 list at the same price as Fable.
Every Claude API price since Claude 3
Prices are per million tokens in US dollars, at each model’s standard list price. Models marked * use the newer tokenizer. “Retired” refers to the Claude API, and some models stay available on cloud platforms for longer. This dataset starts with Claude 3 in March 2024. Download it as a CSV file.
Model
Launched
Input
Output
Status
Source
Claude Haiku 5.5 * Prompts up to 100K tokens. Above that: $0.50 / $2.50.
Launch dates are those on Anthropic’s announcements and release notes. Claude 3.5 Haiku’s date and launch price come from TechCrunch. Status comes from Anthropic’s model deprecations page, and current prices from its pricing page.
When Anthropic changed a price after launch
Anthropic usually changes prices with a new model rather than repricing an existing one. We found these exceptions, all of them in the customer’s favor, but there may be others:
December 3, 2024: Claude 3.5 Haiku went from $1 / $5 to $0.80 / $4, about a month after launch.
March 13, 2026: the higher rate for prompts over 200K tokens ended on Opus 4.6 and Sonnet 4.6. On Opus 4.6, that rate had been $10 / $37.50.
August 10, 2026: Sonnet 5’s introductory $2 / $10 became the standard price, cancelling a planned rise to $3 / $15 on September 1.
October 7, 2026: prompt cache reads on Sonnet 5.5 fell from $0.20 to $0.10. Other Sonnet 5.5 prices did not change.
Why list prices overstate the savings since 2026
Anthropic’s pricing page says Claude 4.7 and later models use a newer tokenizer that produces approximately 30% more tokens for the same text, and that the exact increase depends on the content and workload. Its Sonnet 5 announcement puts the range at roughly 1.0 to 1.35 times, depending on the content type. Because you pay per token, a model at the same list price can cost more for the same work.
This table is our own rough estimate, not an Anthropic figure. It multiplies list prices of models on the newer tokenizer by 1.3 so that they compare with older models:
Comparison
Input list price
Like-for-like input price
Change at list price
Change like-for-like
Claude 3 Opus to Opus 5.5
$15 to $4
$15 to about $5.20
-73%
about -65%
Opus 4.6 to Opus 5.5
$5 to $4
$5 to about $5.20
-20%
about +4%
Opus 5 to Opus 5.5, same tokenizer
$5 to $4
No adjustment needed
-20%
-20%
Sonnet 4.6 to Sonnet 5.5
$3 to $2
$3 to about $2.60
-33%
about -13%
Haiku 4.5 to Haiku 5.5, up to 100K tokens
$1 to $0.10
$1 to about $0.13
-90%
about -87%
Haiku 4.5 to Haiku 5.5, over 100K tokens
$1 to $0.50
$1 to about $0.65
-50%
about -35%
Treat these as a guide only, because real costs can land on either side of this estimate:
Measured token counts vary.Simon Willison counted 1.46 times the tokens for the Opus 4.7 system prompt compared with Opus 4.6 and expected Opus 4.7 to cost around 40% more. TokenCost puts the extra cost at 25 to 37% in production.
Cache reads can pull the other way. In its own measurements, Anthropic says Opus 5.5 costs 40% less than Opus 5 on typical workloads at default settings. Its cache reads are 60% cheaper than those on Opus 5.
Agents can use more tokens per task.Stephan Miller, citing Artificial Analysis, reports about $1.20 per benchmark task on Sonnet 4.6 and $2.29 on Sonnet 5, despite the lower list price.
To measure your own case, use Anthropic’s token counting API on your prompts.
Team prices fall: standard to $25 monthly or $20 annually, premium to $125 or $100. First seen in the archived copy of January 29, and not in the copy of January 28.
Ultra appears at $200 per month on the pricing page, and Max is listed at $100 only, so the $200 Max option is gone. First seen in the archived copy of October 31. API release notes dated October 7 already mention Ultra.
Archive dates show when a change was first visible in a saved copy, so the real date may be earlier. Pro shows $20 per month in every copy we could read, from May 2025 to now, which matches its 2023 launch price. We have no readable copy from before May 2025.
Claude usage limit history
Anthropic’s pricing page describes higher plans as multiples of Pro rather than in messages or tokens, so these changes cannot be charted. Anthropic did give TechCrunch rough weekly estimates for Claude Code in July 2025: 40 to 80 hours of Sonnet 4 on Pro, 140 to 280 on the $100 Max plan and 240 to 480 on the $200 Max plan, varying with codebase size and other factors.
Date
Change
Source
Aug 20, 2025
Team and Enterprise can buy extra usage after hitting their limit
Claude Code’s five-hour limits double for Pro, Max, Team and seat-based Enterprise, and a peak-hours reduction ends for Pro and Max. We did not date when that reduction began. API rate limits for Opus also rise.
Ultra is listed with no 5-hour limit on verified accounts and 10x Pro usage per week. The earlier $200 Max option was described as 20x more usage than Pro, and the older pricing page tied that to the 5-hour session, so the two figures are not like for like.
API prices: Anthropic’s pricing page, release notes and launch announcements. One launch price comes from TechCrunch, as marked.
Plan prices: Anthropic’s announcements and press coverage, plus monthly archived copies of its pricing page from the Internet Archive, compared for changes. Copies before May 2025 contained no readable prices.
Limits: Anthropic’s announcements, its release notes and TechCrunch.
What is missing: Models before Claude 3, regional prices, batch discounts, fast mode, Mythos Preview and prices on cloud platforms. Our archive check is monthly, except where a date is given, so a short-lived change could be absent.
You are welcome to use this data. Please credit AI Rockstars and link to this page. If you find an error, tell us through our site.
Claude price history FAQ
Has Claude gotten cheaper?
At list price, yes. Opus fell 73% from 2024 to 2026, and Sonnet 33% in June 2026. After allowing for the newer tokenizer, our rough estimate is about 65% for Opus and 13% for Sonnet. At the very top, Fable lists at $10 / $50, which is 33% below the 2024 Opus price. Real bills can move differently, as the tokenizer section shows.
Has the price of Claude Pro changed?
Not as far as we can see. It launched at $20 per month in September 2023 and costs $20 today, and every archived copy we could read since May 2025 shows the same price. Annual billing brings it to $17 per month.
Why does my Claude bill not match the price drop?
One possible reason is the newer tokenizer, which produces about 30% more tokens for the same text on Claude 4.7 and later models. Your model mix, caching and how much output you generate also matter.
Which Claude model was the most expensive?
In this dataset, Claude 3 Opus, Opus 4 and Opus 4.1 share the highest base list price at $15 / $75 per million tokens.
Florian Schröder is a co-founder of AI Rockstars and an expert in AI strategies for marketing. He advises companies on integrating AI tools such as ChatGPT, Claude, and Gemini into their workflows. With over 10 years of experience in digital marketing, he combines technical expertise with practical applications.
To provide the best experiences, we use technologies like cookies to store and/or access device information. Consenting to these technologies will allow us to process data such as browsing behavior or unique IDs on this site. Not consenting or withdrawing consent, may adversely affect certain features and functions.
Functional
Always active
The technical storage or access is strictly necessary for the legitimate purpose of enabling the use of a specific service explicitly requested by the subscriber or user, or for the sole purpose of carrying out the transmission of a communication over an electronic communications network.
Preferences
The technical storage or access is necessary for the legitimate purpose of storing preferences that are not requested by the subscriber or user.
Statistics
The technical storage or access that is used exclusively for statistical purposes.The technical storage or access that is used exclusively for anonymous statistical purposes. Without a subpoena, voluntary compliance on the part of your Internet Service Provider, or additional records from a third party, information stored or retrieved for this purpose alone cannot usually be used to identify you.
Marketing
The technical storage or access is required to create user profiles to send advertising, or to track the user on a website or across several websites for similar marketing purposes.
