Claude Sonnet 5: Permanent $2/$10 Pricing and Developer Impact
Anthropic made Claude Sonnet 5 API pricing permanent at $2 per million input and $10 per million output tokens. See a cost calculation with cache assumptions.

Anthropic made Claude Sonnet 5 API pricing permanent at $2 per million input and $10 per million output tokens. See a cost calculation with cache assumptions.
Written and checked by the RankLLMs editorial team. See theeditorial policy.
Anthropic’s August 10, 2026 changelog says Claude Sonnet 5’s introductory API rates of $2 per million input tokens and $10 per million output tokens are permanent; the planned September 1 change to $3/$15 no longer applies. This update uses Anthropic’s release and pricing notice as the source of the historical change. Check the live rate card before budgeting because provider prices and cache terms can change.
Current rates in the announcement
Anthropic lists standard input/output at $2/$10 per million tokens. It also describes prompt caching savings of up to 90%, but the effective rate depends on cache writes, cache reads, and cache lifetime. Do not treat cached input as free or apply the cache-read rate to every prompt token.
The change from planned $3/$15 rates is a one-third reduction in each base rate: (3 - 2) / 3 = 33.3% for input, and (15 - 10) / 15 = 33.3% for output. The comparison is with the announced future rates, not a claim that every workload’s total bill falls by one-third; usage mix and caching matter.
Reproducible cost example
Assume 100,000 reviews per month, each with 120,000 input tokens and 2,500 output tokens. Total monthly volume is 12 billion input tokens and 250 million output tokens. If 80% of input tokens are cache reads, and if cache reads are billed at $0.20 per million as shown in Anthropic’s pricing schedule at the time of the calculation, the hypothetical bill is:
| Token category | Volume | Rate | Cost |
|---|---|---|---|
| Uncached input | 2,400M | $2.00/M | $4,800 |
| Cached input | 9,600M | $0.20/M | $1,920 |
| Output | 250M | $10.00/M | $2,500 |
| Total | $9,220/month |
At the formerly planned base rates ($3 input, $15 output) and the corresponding $0.30 cache-read rate, the same token volumes would total $13,830/month. The modeled difference is $4,610/month, or $55,320 over twelve months if volume and rates stayed constant. This is arithmetic for the stated assumptions; it excludes cache writes, tools, retries, taxes, and other charges. Recalculate using the rates and usage returned for your actual account.
What can change the bill
Coding agents repeatedly send instructions and repository context, so cache behavior can materially affect costs. Cache writes are billed separately, and a changed prompt prefix may reduce reuse. Output tokens can also dominate a task when the model emits long plans or patches. Track cached and uncached input separately instead of assuming every repeated context is a cache hit.
Tokenization can affect usage because providers may count the same text differently. Compare the token counts returned for the exact prompts your application sends, and include those counts in cost estimates.
A subscription is another billing route. Claude Code subscription usage and API metering should not be compared as though they use the same quota. Record the account tier and rate source alongside token counts.
Developer impact
A permanent base rate makes budget estimates easier to maintain than a temporary introductory price, but it does not guarantee lower cost per accepted task. Include retry count, tests that pass, reviewer corrections, and cache-write volume in a pilot. If a rate or model changes, rerun the calculation from raw usage rather than carrying forward the former forecast.
For a matched model comparison, use the same task set and agent harness, then compare total cost and review time. Explore the Claude Sonnet 5 profile, Claude model directory, and LLM comparison tool.
Frequently Asked Questions
What is Claude Sonnet 5's permanent API price?
Claude Sonnet 5 remains permanently at $2.00 per million input tokens and $10.00 per million output tokens.
Was Claude Sonnet 5 originally scheduled to become more expensive?
Yes. Anthropic originally announced that Sonnet 5 would revert to $3.00 per million input and $15.00 per million output tokens after August 31, 2026.
How does Sonnet 5 tokenization impact real costs?
This article does not claim a specific tokenizer-efficiency advantage. Count tokens from the exact prompts used in your application.
Does prompt caching reduce Sonnet 5 costs further?
Yes. Prompt cache reads are billed at $0.20 per million tokens (a 90% discount on input rates).
Source links for data and claims discussed in this article are listed below. Refer to each source for its current figures and methodology.
- •Anthropic Official Platform Pricing(source)
- •Claude API Developer Documentation(source)
Recommended Reading
Related by topic and recency.
Alibaba Model Studio Token Plan: Pricing Structure and Claude Code Integration
Review of Alibaba Model Studio Token Plan terms, Singapore availability, Personal and Team quotas, API compatibility, and billing checks before subscribing.
Claude 3.5 Sonnet Benchmarks: June vs Oct 2024
Compare both 2024 Claude 3.5 Sonnet versions: SWE-bench Verified 33.4% (June) vs 49.0% (Oct), TAU-bench gains, launch pricing, and how to cite the scores.
Best LLMs for Coding in 2026: How to Compare Them
Choose a coding LLM by matching the benchmark to your workflow, checking the agent harness, and measuring task quality and cost on your own repository.
