How Much Does Claude Sonnet 5 Cost?
Claude Sonnet 5 costs $3 per million input tokens and $15 per million output tokens on the Anthropic API. The lower $2/$10 rate quoted almost everywhere online was a launch rate that applied through 31 August 2026 only.
Anthropic published both numbers in the same place. The Sonnet 5 release announcement of 30 June 2026 set the model at $2 per million input and $10 per million output through 31 August 2026, and stated the rate would move to $3 and $15 after that date. That date has passed, so $3/$15 is the figure to build a budget on.
This is the single most commonly wrong detail in Sonnet 5 write-ups. Anything published in July or August quoted the launch rate as though it were standard, and very little of it was revisited on 1 September.
Sonnet 5 is also the default model on the Claude Free plan, the rare current-generation model you can use at no cost in a chat window while paying full metered rates for the same model through the API.
AI Perks tracks the free credit programs that offset that API bill.
The Launch Rate Is Over, and What That Costs
Moving from $2/$10 to $3/$15 is a flat 50% increase on both sides of the bill. No change to your input and output mix softens it.
| Monthly volume | Launch rate $2/$10 | Current rate $3/$15 | Annual difference |
|---|---|---|---|
| 5M in / 1M out | $20 | $30 | $120 |
| 25M in / 5M out | $100 | $150 | $600 |
| 50M in / 10M out | $200 | $300 | $1,200 |
| 200M in / 40M out | $800 | $1,200 | $4,800 |
All figures are estimates, calculated by multiplying the published per-million rates by the volumes shown.
If your cost model still carries the launch rate, every line in it is a third too low. On the heaviest profile above, the correction is $4,800 a year on a single workload.
Sonnet 5 Against the Rest of the Claude Line
At $3/$15, Sonnet 5 costs 60% of Opus 5 and 30% of Fable 5 on identical traffic.
| Model | Released | Input per 1M | Output per 1M | 1M in + 1M out |
|---|---|---|---|---|
| Claude Fable 5 | 9 June 2026 | $10.00 | $50.00 | $60.00 |
| Claude Opus 5 | 24 July 2026 | $5.00 | $25.00 | $30.00 |
| Claude Sonnet 5 | 30 June 2026 | $3.00 | $15.00 | $18.00 |
The blended column is derived from the listed rates at a 1:1 token mix.
Every tier in that table prices output at exactly 5x input. That is not a detail to skip past: because the ratio is identical across tiers, the multiple between any two models stays flat at any traffic shape. Opus 5 costs about 1.67x Sonnet 5 whether your workload is prompt-heavy or generation-heavy, and Fable 5 about 3.33x. The tier decision is the same arithmetic at every mix.
Anthropic positioned Opus 5 as delivering performance close to Fable 5 on many tasks at roughly half the price, built for long-running agentic work, and described it at release as its most-aligned model per its own alignment audit. Fable 5 and Mythos 5 were announced together on 9 June 2026 as a top-tier Mythos-class pair for autonomous knowledge work and coding, with a 1 million token context window and 128K maximum output.
AI Perks tracks which providers currently hand out credits against these rates.
What Sonnet 5 Actually Costs to Run
At 50 million input and 10 million output tokens a month, Sonnet 5 runs about $300. The same workload runs about $500 on Opus 5 and about $1,000 on Fable 5.
| Monthly volume | Sonnet 5 | Opus 5 | Fable 5 |
|---|---|---|---|
| 5M in / 1M out | $30 | $50 | $100 |
| 25M in / 5M out | $150 | $250 | $500 |
| 50M in / 10M out | $300 | $500 | $1,000 |
| 200M in / 40M out | $1,200 | $2,000 | $4,000 |
All figures are estimates derived from the published per-million rates.
Notice what a 5:1 input to output ratio does. Because output is priced at 5x input, those two halves of the bill land exactly equal: at 50M in and 10M out, input costs $150 and output costs $150. Teams who assume input dominates, and who spend their engineering effort compressing prompts, are usually working the cheaper half of the problem.
The strongest lever on a Sonnet 5 bill is shorter responses, not shorter prompts. A change that cuts average output length by 40% moves more money than most prompt compression work, and it is usually easier to ship.
The Two Discounts Anthropic Publishes
On Opus 5, Anthropic publishes a Batch API rate at exactly half of standard and a prompt cache read rate at one tenth of standard input. Those are the two levers to pull before any other cost work.
| Opus 5 billing mode | Input per 1M | Output per 1M |
|---|---|---|
| Standard API | $5.00 | $25.00 |
| Batch API | $2.50 | $12.50 |
| Prompt cache read | $0.50 | not applicable |
Opus 5 also carries a 1.1x multiplier for US-only inference, which is the line that quietly raises a bill for teams carrying a data residency requirement.
The shape of the discount is what transfers: batch work that does not need an answer within seconds runs far cheaper, and a long system prompt or reference document resent on every call belongs in the cache rather than in the request body. Sonnet 5 batch and cache rates are not asserted here, so confirm the current ones on Anthropic's pricing page before you model them.
Is Claude Sonnet 5 Free?
Yes in the Claude app, no on the API. Sonnet 5 is the default model on the Claude Free plan and is available on Pro, Max, Team, Enterprise and Claude Code. Every API token is metered from the first call.
That split catches teams out constantly. You can use Sonnet 5 all day in the chat interface at no cost, then find that wiring the same model into your product starts a real bill at $3/$15.
The higher tiers have no free path at all. Opus 5 is not included on the free plan, and Fable 5 is a premium-tier model with no published free tier, so anything above Sonnet means a paid subscription or metered API usage.
Which is why free API credits are worth more on Anthropic than on most providers. Anthropic runs Claude for Startups, which provides free API credits, priority rate limits and early access to new models. Amounts and current conditions are the part that moves, and AI Perks tracks what is open.
Covering the Sonnet 5 Bill With Credits
A $300 a month Sonnet 5 workload is $3,600 a year. Credit programs can cover a meaningful share of a first year, which turns the model choice from a budget question into a purely technical one.
Three things are worth knowing before you count on them.
Model credits and compute credits are separate bills. Anthropic API credits and cloud compute credits come from different programs on different terms, and teams holding both cover far more of a first year than teams holding one.
Terms move faster than articles do. The jump from $2/$10 to $3/$15 on 1 September is the proof. A price someone researched in July was a different price by September, and credit amounts and program conditions move on the same kind of timetable. A live tracker beats an article read once, which is what AI Perks exists to be.
Batching and caching stack with credits. They are not alternatives to each other. A workload that is heavily batched and heavily cached burns a credit balance far more slowly, which stretches the same allocation across more months of runway.
Current Anthropic programs, alongside the rest of the AI and infrastructure providers offering credits, are listed at getaiperks.com.
Frequently Asked Questions
How much does the Claude Sonnet 5 API cost?
Claude Sonnet 5 costs $3 per million input tokens and $15 per million output tokens. The $2/$10 rate widely quoted online was the launch rate announced with the 30 June 2026 release, which Anthropic stated would apply through 31 August 2026.
Did Claude Sonnet 5 get a price increase in September 2026?
Yes. Anthropic set Sonnet 5 at $2 per million input and $10 per million output through 31 August 2026, and published $3 and $15 as the rates from after that date. Any article describing $2/$10 as the standard price predates 1 September 2026 and is out of date.
Is Claude Sonnet 5 free to use?
Sonnet 5 is the default model on the Claude Free plan, so it is free in the chat app, and it is available on Pro, Max, Team, Enterprise and Claude Code. API access is metered from the first token at $3/$15 per million. Free API credits are the way to cover that, and AI Perks tracks which programs are open.
Is Claude Sonnet 5 cheaper than Claude Opus 5?
Yes, by roughly 1.67x on both input and output. Opus 5 is $5 per million input and $25 output, against Sonnet 5 at $3 and $15. Because both models price output at 5x input, that multiple holds at every possible input to output mix.
How does Sonnet 5 compare with Claude Fable 5?
Fable 5, announced on 9 June 2026 alongside Mythos 5, is the top-tier Mythos-class model built for autonomous knowledge work and coding, with a 1 million token context window and 128K maximum output. It is priced at $10 per million input and $50 per million output, roughly 3.33x Sonnet 5 on identical traffic.
How do I get free Claude API credits?
Anthropic runs Claude for Startups, which provides free API credits, priority rate limits and early access to new models. Credits apply to Sonnet 5 like any other model, so a $3,600 annual API bill can drop substantially. Terms change often, so check which programs are currently open at getaiperks.com.
The launch rate is gone. The credits are not.