Sonnet 5.5 Price: Same $2/$10 as Sonnet 5, and Whether to Switch

Data checked: September 29, 2026. Specs and prices come from the BazaarLink public model catalog (/api/v1/models). Benchmark scores come from Artificial Analysis (AA below). Both were checked on the same day.

Claude Sonnet 5.5 (model ID claude-sonnet-5.5) is now available on BazaarLink.

Price data as of 2026-10-09, checked against the BazaarLink model list; the rates match this article. Actual BazaarLink billing follows the model page and your invoice.

Claude Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens, with a 1,000,000-token context window. That is the same as Sonnet 5 and half the rate of Opus 5.5 ($4 and $20). Prices are in US dollars. See the live price and parameters on the Sonnet 5.5 model page, and other models at /en/models.

The short version

• The price did not change. Sonnet 5.5 and Sonnet 5 both cost $2 per million input tokens and $10 per million output tokens, with a 1,000,000-token context at one flat price and no long-context surcharge. Switching does not raise your unit cost.
• The AA Intelligence Index gap is large. At maximum reasoning effort (Max), Sonnet 5.5 scores 56 and Sonnet 5 scores 38.2. At the default high effort it is 46.7 versus 31.7. This is one benchmark suite. It does not mean your own tasks will improve by the same amount.
• One behavior difference to note: the public catalog marks reasoning as mandatory for Sonnet 5.5 (mandatory: true) and not mandatory for Sonnet 5. According to Anthropic's docs, sending thinking: disabled to Sonnet 5.5 returns a 400, and up-front thinking is turned off with between_tools (high effort or below). If your workflow depends on switching reasoning off, rewrite and test before you switch. See what Anthropic officially says and which API changes to make.
• Because the price is the same, the cost of switching is testing time, not your bill. Run a set of representative tasks side by side before switching everything.

Specs and pricing

Price and context are identical, so there is no "upgrade costs more" question. The BazaarLink catalog describes Sonnet 5.5 as a Sonnet-class model for well-scoped everyday work, succeeding Sonnet 5 as a direct upgrade. Live prices are in the BazaarLink model catalog.

Cost example

Unit prices per million tokens (identical for Sonnet 5 and 5.5; cache prices are for Sonnet 5.5):

A monthly customer-support bot example: 500 conversations a day, 3 turns each, 300 input + 150 output tokens per turn. That is about 13.5 million input and 6.75 million output tokens a month, or 13.5 x $2 + 6.75 x $10 = $94.5. The figure is the same on Sonnet 5 and 5.5. Your real monthly cost depends on your own usage records.

Price compared with Opus 5.5 and Haiku 5.5

This compares published list prices only and ranks nothing on capability. Rates from the BazaarLink model list (2026-10-09):

Assume 1,000 requests a day, 10,000 input and 1,000 output tokens each, over 30 days (300M input and 30M output tokens; plain arithmetic that ignores caching discounts and retries). Sonnet 5.5 comes to about $900 a month, Opus 5.5 about $1,800 and Haiku 5.5 about $45. The Haiku 5.5 figures are the model list rates; per Anthropic's announcement, Haiku 5.5 prompts above 100,000 tokens are $0.50 input and $2.50 output. Actual BazaarLink billing follows the model page and your invoice.

What Artificial Analysis says

AA is an independent model benchmarking site. As of September 29, 2026, the Intelligence Index for Sonnet 5.5 and Sonnet 5 is (scored separately per reasoning effort):

At Max effort, Sonnet 5.5 ranks 2nd of the 670 models AA tracks on the Intelligence Index (Sonnet 5 ranks 62nd). AA records a median output speed of 141.87 tokens per second, against 81.31 for Sonnet 5. Keep in mind:

• Scores are per model and effort combination, so different effort levels cannot be compared directly.
• The Intelligence Index is AA's own benchmark suite. It does not mean your tasks, languages or tool setups will improve by the same amount.
• These figures come from AA's Sonnet 5.5 page and Sonnet 5 page and may be updated.

Who should switch?

Worth trying Sonnet 5.5 first:

• You use Sonnet 5 for everyday coding, document work or support and want to try a higher benchmark score at the same unit price.
• New projects: at the same price there is no pricing reason to pick the older version.

Test before switching if:

• Your workflow depends on turning reasoning off or on very low latency.
• Your prompts and output formats are tuned and you need to confirm the new version's output style does not break downstream parsing.

Suggested process: pick 20 to 30 real tasks, fix the prompts, tools and reasoning effort, run both Sonnet 5 and 5.5, and record pass rate, manual fix time and input/output tokens. Switch everything once it checks out. If you are comparing a higher tier, see Claude Opus 5 vs Opus 5.5.

FAQ

How much does the Claude Sonnet 5.5 API cost per million tokens?

The BazaarLink model list (2026-10-09) shows $2 input and $10 output per million tokens, the same as Sonnet 5. Actual BazaarLink billing follows the model page and your invoice.

Is Sonnet 5.5 more expensive than Sonnet 5?

No. Both cost $2 per million input tokens and $10 per million output tokens, with a 1,000,000-token context. The unit price is identical. Your total cost depends on your input and output volume.

How long is the Sonnet 5.5 context, and does price rise with length?

The context length is 1,000,000 tokens at one flat price, with no long-context tiers.

Can I keep using Sonnet 5?

Yes. Sonnet 5 is still in the BazaarLink model catalog. Since the price is the same, you can test side by side without rushing to switch everything.

Does Sonnet 5.5 accept images and files?

Yes. Input can be text, images and files, and output is text.

How often are the scores and prices in this article updated?

Prices and specs follow the BazaarLink public model catalog and scores follow the AA website. This article was checked on September 29, 2026. Use the live pages for later changes.

TWD billing · Taiwan invoices · leading AI models · OpenAI-compatible API