Claude Sonnet 5 launched on 30 June 2026 at an introductory price of $2 per million input tokens and $10 per million output tokens, valid through 31 August 2026. From 1 September the price was due to rise to $3 / $15, the same as the previous Sonnet generation. In mid-August Anthropic reversed that plan. Its pricing documentation now says the $2 / $10 launch price is the standard price and that the scheduled increase to $3 / $15 will not occur.
If your budget or pricing assumed a 50% rise in Sonnet costs from September, that line can come out. But the per-token rate is only half of the story, because of how Sonnet 5 counts tokens.
The timeline
- 30 June 2026: Sonnet 5 launches at $2 / $10, described as introductory pricing through 31 August.
- Mid-August 2026: Anthropic announces the introductory price is permanent.
- 1 September 2026: passes with no change. Sonnet 5 remains $2 / $10.
The full rate card
- Input: $2 per million tokens.
- Output: $10 per million tokens.
- Cache reads: $0.20 per million tokens, a tenth of the input price.
- Batch processing: 50% off both input and output.
For comparison, Anthropic's other current models per million input / output tokens: Claude Haiku 4.5 at $1 / $5, Claude Opus 5.5 at $4 / $20, Claude Opus 5 at $5 / $25 and Claude Fable 5.1 at $10 / $50. Since 22 September, OpenAI's GPT-6 Sol also lists at exactly $2 / $10, so the two mid-tier models now cost the same per token.
What a typical workload costs
For a request with 5,000 input tokens and 1,000 output tokens, run 10,000 times a month:
Input
5,000 × $2.00 ÷ 1Mequals$0.01000
Output
1,000 × $10.00 ÷ 1Mequals$0.01000
Per month
$0.02000 × 10,000equals$200.00
That is $200 a month. Had the planned increase gone ahead, the same workload would have cost $300.
Why the real saving is smaller than it looks
Sonnet 5 uses a new tokenizer, and independent reviewers report that it turns the same text into roughly 30% more tokens than the previous generation. Anthropic set the introductory price partly with this in mind, so that moving to Sonnet 5 would stay roughly cost-neutral. It means you cannot compare Sonnet 5 and Sonnet 4.6 on the rate card alone.
Here is what that does to a single request. Suppose a prompt was 4,000 input tokens with a 1,000-token answer on Sonnet 4.6 at $3 / $15. That cost $0.027. If the same text comes to about 30% more tokens on Sonnet 5 (5,200 in and 1,300 out) at $2 / $10, it costs about $0.0234. That is about 13% cheaper per request, not the 33% the rate card suggests. The exact figure depends on your text: code, non-English languages and structured data can tokenize differently.
What to do now
- Re-forecast any budget built on $3 / $15 from September. Use $2 / $10 as the standard rate.
- Before migrating from an older model, count tokens on a sample of your real prompts with the new model, rather than assuming the old counts carry over.
- Turn on prompt caching for repeated system prompts and reference material. At $0.20 per million, cached input is a tenth of the normal price.
- Move work that can wait (bulk classification, overnight summaries) to batch processing at half price.
- If you parked a migration to Sonnet 5 because of the expected price rise, it is worth revisiting.
Tip: Put your real token counts into the AI API Cost Calculator and compare Sonnet 5 with the model you use today. Measure tokens on the new model, not the old one.
Questions people ask
- How much does Claude Sonnet 5 cost?
- $2 per million input tokens and $10 per million output tokens, now the standard permanent rate. Cache reads are $0.20 per million and batch processing is half price.
- Did Claude Sonnet 5's price go up in September 2026?
- No. The planned increase to $3 and $15 on September 1, 2026 was cancelled, and Anthropic's pricing docs say it will not occur.
- Is Claude Sonnet 5 cheaper than Sonnet 4.6?
- Per token, yes: $2 / $10 against $3 / $15. But Sonnet 5's tokenizer produces roughly 30% more tokens for the same text, so the saving per request is closer to 13% in a typical example.
- How does Sonnet 5 compare with other Claude models on price?
- Haiku 4.5 is $1 and $5, Opus 5.5 is $4 and $20, Opus 5 is $5 and $25, and Fable 5.1 is $10 and $50 per million input and output tokens.
- Is Claude Sonnet 5 the same price as GPT-6 Sol?
- Yes, both list at $2 per million input tokens and $10 per million output tokens. Real costs differ with how many tokens each model uses for your task.
Sources
Get fee-change alerts
One short email when a price our calculators use changes. No other mail, unsubscribe anytime.