What Does the Claude API Cost in 2026?
Anthropic's current line runs from $1 per million input tokens at the bottom to $10 at the top, and every model in it prices output at exactly five times input.
All figures below are USD per million tokens, input / output, checked against Anthropic's own pricing documentation on 16 September 2026.
| Model | Input / 1M | Output / 1M | 1M in + 1M out | Notes |
|---|---|---|---|---|
| Claude Fable 5.1 | $10.00 | $50.00 | $60.00 | Top tier, Mythos-class, released September 2026 |
| Claude Mythos 5.1 | $10.00 | $50.00 | $60.00 | Limited availability |
| Claude Opus 5 | $5.00 | $25.00 | $30.00 | Launched 24 July 2026 |
| Claude Sonnet 5 | $2.00 | $10.00 | $12.00 | Scheduled increase cancelled |
| Claude Haiku 4.5 | $1.00 | $5.00 | $6.00 | Cheapest current Claude tier |
The fourth column is arithmetic on the two published prices, not a figure Anthropic publishes. It is the number worth reasoning with, because almost no real workload is input-only.
Two things in that table are easy to miss. Opus 5 launched at the same price as Opus 4.8, so a new generation of that tier arrived with no price increase attached. And the five-times output ratio holds across all four price points, which makes Claude cost modelling unusually simple: work out your output tokens, multiply by five, add your input tokens, and you have your bill in input-equivalents.
Credits that cover these bills outright are tracked at AI Perks, which currently follows $7.7M in credits across 194 companies.
The Sonnet 5 Price Increase Anthropic Cancelled
Sonnet 5 was scheduled to rise from $2/$10 to $3/$15 per million tokens on 1 September 2026. Anthropic cancelled that increase, and $2/$10 is now the standard price rather than an introductory one.
This is the most under-reported number on this page. The $2/$10 rate was originally announced as introductory through 31 August 2026, which means a great deal of budgeting advice written earlier in the year assumes you are now paying $3/$15.
Anthropic's pricing page states it plainly: "The previously scheduled increase to $3/$15 per million input/output tokens on September 1, 2026 will not occur."
What that is worth, using the listed prices:
| Workload per month | At $3/$15 (cancelled) | At $2/$10 (actual) | Difference |
|---|---|---|---|
| 10M in / 2M out | $60 | $40 | $20 |
| 50M in / 10M out | $300 | $200 | $100 |
| 250M in / 50M out | $1,500 | $1,000 | $500 |
Those are estimates derived from list prices, before any discounts. The ratio is constant: the cancellation holds Sonnet 5 a third below what was scheduled, whatever your volume.
Claude vs OpenAI: The Same Table, Both Vendors
At the top of each line the two vendors are priced identically. Claude Fable 5.1 and gpt-6-astra both list $10 input and $50 output per million tokens.
OpenAI's current line, checked the same day:
| Model | Input / 1M | Output / 1M | Output multiple |
|---|---|---|---|
| gpt-6-astra | $10.00 | $50.00 | 5x |
| gpt-5.6-sol | $4.00 | $20.00 | 5x |
| gpt-5.6-terra | $2.00 | $12.00 | 6x |
| gpt-5.6-luna | $0.20 | $1.20 | 6x |
| gpt-5.5 | $5.00 | $30.00 | 6x |
| gpt-5.4 | $2.50 | $15.00 | 6x |
| gpt-5.4-mini | $0.75 | $4.50 | 6x |
| gpt-5.4-nano | $0.20 | $1.25 | 6.25x |
gpt-6-astra launched 3 to 4 September 2026 and carries a separate long context rate above 200,000 tokens, at $20.00 / $75.00. No equivalent surcharge appears in the Anthropic prices checked here. The Sol, Terra and Luna names replaced the older Instant, Thinking, mini and nano scheme in July 2026.
Three comparisons that actually decide things:
Flagship for flagship, it is a draw on list price. $10/$50 either way, before OpenAI's long context surcharge above 200,000 tokens, which has no counterpart in the Anthropic prices checked here.
In the middle, Claude is slightly cheaper per job. Sonnet 5 costs $12 per million tokens each way. gpt-5.6-terra, at the same $2 input, costs $14. gpt-5.4 costs $17.50. All three are derived by adding the listed input and output prices.
At the bottom, Claude is not competitive on price. Haiku 4.5 at $6 per million each way sits at more than four times the $1.40 that gpt-5.6-luna costs on the same basis. Both figures are derived from list prices.
One warning: OpenAI's legacy gpt-5, gpt-5-mini and gpt-5-nano are cheap on paper at $1.25/$10, $0.25/$2 and $0.05/$0.40, but that API shuts down on 10 December 2026. Do not start anything new on them. If you want free credits to migrate off them, AI Perks tracks what each vendor is handing out.
Which Claude Tier Fits Your Workload?
Pick by output volume, not by benchmark reputation, because a single output token costs five times what an input token costs on every Claude model.
Here is a moderate production workload of 50M input and 10M output tokens per month, costed from the listed prices. Treat every figure as an estimate.
| Model | Monthly estimate | Cost shape on this workload |
|---|---|---|
| Claude Haiku 4.5 | $100 | The floor of the current line, so it absorbs the most calls per budget dollar |
| Claude Sonnet 5 | $200 | Twice Haiku 4.5, half Opus 5 |
| Claude Opus 5 | $500 | Five times Haiku 4.5 on identical traffic |
| Claude Fable 5.1 | $1,000 | Ten times Haiku 4.5, the ceiling of the current line |
The spread from Haiku to Fable is ten times on identical traffic. That is the entire argument for routing: send the cheap, structured, high-volume calls to Haiku 4.5 and reserve Opus 5 or Fable 5.1 for the calls where a wrong answer is expensive. A mixed deployment that puts 80 percent of traffic on Haiku and 20 percent on Opus lands near $180 per month on this workload, cheaper than running everything on Sonnet 5.
Anthropic has not published capability rankings we can verify here, so this table is about cost shape only. AI Perks is where to look for credits that flatten the difference.
What the Other Providers Charge
Meta's Muse Spark line is the only other provider with prices verified in this research, at $1.25 input and $4.25 output per million tokens.
| Provider and model | Input / 1M | Output / 1M |
|---|---|---|
| Meta Muse Spark 1.1 / 1.2 / 1.3 | $1.25 | $4.25 |
| Meta Muse Spark, Contributor tier | $0.10 | $0.20 |
| Meta Muse Image | $0.01 per image | n/a |
The Contributor tier is the cheapest per-token price on this page, and the reason is stated by Meta: on that tier, Meta trains on your prompts and completions. For anything touching customer data, that is not a pricing decision, it is a data governance one. Meta also runs no startup credit program for Muse, so there is no free tier to fall back on.
Two deliberate omissions, both because guessing is worse than silence:
xAI Grok. Token prices vary by model and the sources available conflict badly. Check xAI's own console for current rates. We will not print a number here.
Google Gemini. Not verified in this round of research, so it is left out of every table above rather than estimated from memory.
How to Cover the Bill With Credits
Every price on this page is a list price, and list prices are what you pay when you have not applied for credits.
The gap is large. A team running the 50M / 10M workload above on Sonnet 5 spends $2,400 a year at list, and $6,000 on Opus 5. Both are estimates derived from the listed prices, twelve times the monthly figures above. Credit programs across the major vendors routinely cover that entire first year and then some, which is why the cheapest model is often not the right question.
AI Perks tracks $7.7M in credits across 194 companies, including the current Anthropic terms, in one place that stays current as programs change. The price tables above tell you what a workload costs. The credits tell you whether you need to pay for it at all.
Get the credit list at getaiperks.com →
Frequently Asked Questions
How much does the Claude API cost per million tokens in 2026?
Claude Fable 5.1 and Mythos 5.1 cost $10 input and $50 output per million tokens. Opus 5 is $5 and $25, Sonnet 5 is $2 and $10, and Haiku 4.5 is $1 and $5. Every tier prices output at five times input. Credits that offset those bills are tracked at getaiperks.com.
Did Anthropic raise Sonnet 5 pricing in September 2026?
No. The increase from $2/$10 to $3/$15 per million tokens was scheduled for 1 September 2026 and then cancelled. Anthropic's pricing page states the increase will not occur, which makes $2/$10 the standard Sonnet 5 price rather than an introductory rate that expired on 31 August.
Is Claude cheaper than GPT-6?
At the top, no. Claude Fable 5.1 and gpt-6-astra both list $10 input and $50 output per million tokens. Below that it varies by tier: Sonnet 5 at $2/$10 undercuts gpt-5.6-terra at $2/$12, while gpt-5.6-luna at $0.20/$1.20 is far cheaper than Haiku 4.5 on the same basis.
Which Claude model is cheapest?
Claude Haiku 4.5 at $1 input and $5 output per million tokens is the cheapest current Claude model, roughly $6 for a million tokens each way. It is not the cheapest model on the market, though: OpenAI's gpt-5.6-luna lands near $1.40 for the same combination, an estimate based on listed prices.
Does Anthropic charge extra for long context?
OpenAI publishes a long context rate for gpt-6-astra above 200,000 tokens, at $20 input and $75 output per million. No equivalent surcharge appears in the Anthropic prices checked here, so treat the listed Claude rates as flat until Anthropic publishes otherwise. Verify before budgeting a long context workload.
How do I get free Anthropic credits?
Anthropic and most major AI vendors run credit programs that cover API spend outright, which changes the maths on every price in this table. AI Perks tracks $7.7M in credits across 194 companies, including the current Anthropic terms, at getaiperks.com. Programs change terms often, so re-check before you budget.
Every number here was checked on 16 September 2026. Check the credits before you check out.