---
title: "Claude Sonnet 5 is still $2/$10: correcting our false 50% price-increase report"
description: "Anthropic live pricing lists Claude Sonnet 5 at $2/$10 per MTok. We correct our false $3/$15 report and recompute standard, cached, and Batch workloads."
canonical: https://vynaris.com/blog/sonnet-5-price-increase-3-15-september-2026
author: "Dhrumil Barot"
author_bio: "Dhrumil is an AI inference and LLM infrastructure expert. Dhrumil built Vynaris to make model access, routing and pricing simple and transparent for developers."
published: 2026-09-13
updated: 2026-09-13
---

# Claude Sonnet 5 is still $2/$10: correcting our false 50% price-increase report

Published 2026-09-13 · By [Dhrumil Barot](https://vynaris.com/about#founder), Founder

Anthropic live pricing lists Claude Sonnet 5 at $2/$10 per MTok. We correct our false $3/$15 report and recompute standard, cached, and Batch workloads.

**Updated 2026-09-13: correction.**  [Claude Sonnet 5](https://vynaris.com/models#claude-sonnet-5) still costs $2 per million input tokens and $10 per million output tokens. Our previous $3/$15 report was wrong. A 10k-input, 2k-output call costs $0.04, not $0.06. Prices verified 2026-09-13.

## TL;DR

- Anthropic's live page lists $2/$10 as the standard Sonnet 5 price. It explicitly says the scheduled $3/$15 increase did not occur.
- The correction cuts our 10k-input, 2k-output estimate from $0.06 to $0.04 per call. That is $40, not $60, per 1,000 calls.
- Cache reads cost $0.20 per MTok. Five-minute cache writes cost $2.50, one-hour writes cost $4, and Batch costs $1/$5.
- We found no first-party support for our previous claim that the lower price ended. This update replaces that claim instead of inventing a new price event.

## Corrected verdict table

| Line item | Wrong figure previously shown | Live Anthropic price | Correction |
| --- | --- | --- | --- |
| Input per MTok | $3 | $2 | 33.3% lower |
| Output per MTok | $15 | $10 | 33.3% lower |
| Cache read per MTok | $0.30 | $0.20 | 33.3% lower |
| Cache write, 5 minutes | $3.75 | $2.50 | 33.3% lower |
| Cache write, 1 hour | $6 | $4 | 33.3% lower |
| Batch input per MTok | $1.50 | $1 | 33.3% lower |
| Batch output per MTok | $7.50 | $5 | 33.3% lower |
| 1,000 standard calls, 10k in + 2k out | $60 | $40 | $20 lower |
| 1,000 cache-heavy calls | $38.40 | $25.60 | $12.80 lower |
| 1,000 Batch calls, 10k in + 2k out | $30 | $20 | $10 lower |

The primary receipt is [Anthropic's pricing documentation](https://platform.claude.com/docs/en/about-claude/pricing). Its Sonnet 5 row shows $2 input, $2.50 for a five-minute cache write, $4 for a one-hour cache write, $0.20 for a cache hit, and $10 output per MTok.

The note under that table is unusually clear. Anthropic says the $2/$10 launch price is now the standard price. It also says the previously scheduled September 1 increase to $3/$15 will not occur. The live page does not support the opposite interpretation.

## The corrected $0.04 call

Our representative call has 10,000 [input tokens](https://vynaris.com/glossary/input-token-cost) and 2,000 [output tokens](https://vynaris.com/glossary/output-token-cost). The calculation uses only the live list prices:

`10,000 x $2 / 1,000,000 + 2,000 x $10 / 1,000,000 = $0.02 + $0.02 = $0.04`

Across 1,000 calls, the same workload consumes 10 million input tokens and 2 million output tokens:

`10M x $2/MTok + 2M x $10/MTok = $20 + $20 = $40`

Our false $3/$15 estimate produced $60 for the same work. The corrected bill is $20 lower. Dividing the $20 correction by the false $60 estimate gives 33.3%.

Change the token shape in the [cost calculator](https://vynaris.com/calculator?in=10000&out=2000) when your prompts or responses differ. The price correction stays proportional, but the input and output shares change.

## Caching is $0.20 per MTok, not $0.30

The live [prompt caching](https://vynaris.com/glossary/prompt-caching) rates are $2.50 per MTok for a five-minute write, $4 for a one-hour write, and $0.20 for a cache read. These equal 1.25 times, 2 times, and 0.1 times the $2 base input rate.

Take 1,000 calls with 2,000 fresh input tokens, 8,000 cached input tokens, and 2,000 output tokens each. The corrected bill is:

`2M x $2 + 8M x $0.20 + 2M x $10 = $4 + $1.60 + $20 = $25.60`

The article previously showed $38.40 using the false $3 base price. The live-rate total is $12.80 lower. Our [prompt-caching break-even analysis](https://vynaris.com/blog/prompt-caching-break-even-2026) explains why the read-count threshold does not change: all Sonnet 5 cache prices remain fixed multiples of base input price. The dollar bill changes; the multiplier math does not.

## Batch remains $1/$5

Anthropic states that the [Batch API](https://vynaris.com/glossary/batch-api) gives a 50% discount on both input and output tokens. Half of $2/$10 is $1/$5.

For the same 1,000-call workload, Batch costs:

`10M x $1/MTok + 2M x $5/MTok = $10 + $10 = $20`

The false $3/$15 base would have implied $30 after the same discount. The correction removes $10 from that estimate. Our [Batch API comparison](https://vynaris.com/blog/batch-apis-openai-vs-anthropic-when-latency-is-free) remains useful for the latency tradeoff, but any Sonnet 5 line using $1.50/$7.50 needs the same pricing correction.

## What we got wrong

The prior version of this article said Anthropic had moved Sonnet 5 to $3/$15. We also claimed that $2/$10 had lasted only through August 31. A fresh fetch of the first-party page disproves both claims.

The current page preserves the original promotion dates as context, then says the launch price became permanent. We treated the scheduled increase as the current rate. That was a source-reading failure, not an ambiguous calculation.

Our earlier [August pricing update](https://vynaris.com/blog/sonnet-5-intro-pricing-ends-aug-31) had already reported that Anthropic canceled the increase. The later article should not have reversed that conclusion without stronger first-party evidence. This correction restores the live price and removes every downstream claim built on $3/$15.

## Model identity is not ambiguous

Anthropic's [current models overview](https://platform.claude.com/docs/en/about-claude/models/overview) lists the active API model ID as `claude-sonnet-5`. The same first-party model record carries $2 input and $10 output per MTok. We are not mixing Sonnet 5 with [Claude Sonnet 4.6](https://vynaris.com/models#claude-sonnet-4-6), a partner-platform price, or a retired snapshot.

This distinction matters because model aliases can move while a dated model ID stays fixed. Here, both the model catalog and pricing page agree on the active name and the $2/$10 rates.

## Honest tradeoff: a lower price is not a routing verdict

The corrected price makes Sonnet 5 cheaper than our article claimed. It does not prove that every request belongs on Sonnet 5. A cheaper model can still win simple extraction or classification work if it meets the same quality bar. A more capable model can be worth a higher [per-request cost](https://vynaris.com/glossary/per-request-cost) when a failure triggers expensive review or retries.

Do not change a [model routing](https://vynaris.com/glossary/model-routing) policy because one article corrected its arithmetic. Update the rate card, rerun the task-level evaluation, and compare cost per successful outcome. The useful result is not that Sonnet 5 is cheap. It is that its current standard API price is $2/$10, and budgets should use that figure.

## FAQ

**What does Claude Sonnet 5 cost?**

The standard Claude API price is $2 per million input tokens and $10 per million output tokens. Prices were verified on 2026-09-13.

**Did the price rise to $3/$15 on September 1?**

No. Anthropic's current page says the previously scheduled increase would not occur. Our earlier statement that it did occur was false.

**What does a 10k-input, 2k-output call cost?**

It costs $0.04: $0.02 for input plus $0.02 for output. One thousand identical calls cost $40.

**What are the cache and Batch prices?**

Cache reads cost $0.20 per MTok. Cache writes cost $2.50 for five minutes or $4 for one hour. Batch input and output cost $1/$5 per MTok.

**Should I use Sonnet 5 for every task now?**

No. Price is one input to a routing decision. Evaluate accuracy, latency, retries, and review cost on your workload before changing production traffic.

## Sources

- [Anthropic API pricing](https://platform.claude.com/docs/en/about-claude/pricing), fetched and verified 2026-09-13.
- [Anthropic models overview](https://platform.claude.com/docs/en/about-claude/models/overview), model name and API ID verified 2026-09-13.
- Reproducible arithmetic: `artifacts/sonnet-5-price-increase-3-15-september-2026-math.py`.

## About the author

[Dhrumil Barot](https://vynaris.com/about#founder), Founder. Dhrumil is an AI inference and LLM infrastructure expert. Dhrumil built Vynaris to make model access, routing and pricing simple and transparent for developers. [LinkedIn](https://www.linkedin.com/in/bdhrumil) · [X](https://x.com/fyidhrumil)