What Is Claude Sonnet 5.5? Benchmarks, Cost, Free Plan

Anthropic released Claude Sonnet 5.5 on September 28, 2026. It is the second model in the Claude 5.5 family, after Opus 5.5. According to Anthropic's announcement it runs more than 30% faster than Sonnet 5 and costs up to 30% less for most work. Its jump in agentic coding is far beyond Sonnet 5.

This article summarizes the Sonnet 5.5 section of the AI Search channel's weekly AI news video on YouTube and extends it with Anthropic's announcement page, its price table and independent measurements; the video link is at the end. This is a news roundup, and I have not tested the model. Numbers come from Anthropic's page and from the Artificial Analysis charts shown in the video.

What is Claude Sonnet 5.5?

The new version of Anthropic's mid-size model line. Its bigger sibling Opus 5.5 is built for complex work that needs judgment; Sonnet 5.5 is described as strongest at well-scoped everyday tasks, fixing bugs, and producing polished documents, slides and spreadsheets. Anthropic also stresses a good eye for design. The smallest member of the family, Claude Haiku 5.5, arrives in the coming weeks.

At a glance:

Item	Value
Release date	September 28, 2026
API model ID	claude-sonnet-5-5
Input price	$2 per million tokens
Output price	$10 per million tokens
Cache read / write	$0.20 / $2.50 (per million tokens)
Speed	30%+ faster than Sonnet 5
Platforms	Claude Platform, Amazon Web Services, Google Cloud, Microsoft Azure
Data retention	Zero data retention option available

Benchmark results

Selected rows from the table on Anthropic's page:

Benchmark	Sonnet 5.5	Sonnet 5	Opus 5.5
Terminal-Bench 4.0 (agentic terminal coding)	70.6%	10.3%	66.4% (Xhigh effort)
CursorBench 4.0 (tasks from real Cursor sessions)	55.5%	34.1%	57.8%
GDPval-AA v2.1 (real work across 44 occupations, Elo)	1844	1449	1846
AA-Briefcase v1.1 (long-horizon knowledge work, Elo)	1811	1359	1822
Humanity's Last Exam (with tools)	64.5%	54.9%	67.7%
OSWorld 2.1 (computer use)	80.1%	57.0%	81.8%

Two details matter. First, on Terminal-Bench 4.0 Sonnet 5.5 is above Opus 5.5's published score. Second, Anthropic does not overstate it: the page says Opus 5.5 remains clearly stronger on complex, open-ended work that needs sustained judgment. So one benchmark score does not mean Sonnet catches Opus on every job.

In the video, the presenter says that on the Artificial Analysis intelligence index Sonnet 5.5 sits in second place, ahead of GPT-6 Astra.

The "small model means cheap" trap

On the price list Sonnet 5.5 looks attractive: exactly half of Opus 5.5's $4 / $20. But the Artificial Analysis charts in the video show something else: when you compute total cost per task, Sonnet 5.5 comes out more expensive than Claude Fable and Claude Opus. The reason is that it produces more output tokens per task than rivals.

This does not contradict Anthropic's "up to 30% cheaper per task than Sonnet 5" claim, because the two measurements compare different things. One compares Sonnet 5.5 with the previous Sonnet, the other with other companies' models and with Opus. The charts on Anthropic's page confirm the point: cost depends heavily on the effort level you choose (Low, Medium, High, Xhigh, Max). At low effort Sonnet 5.5 is cheap and sufficient; at Max it approaches Opus at a similar cost.

The practical conclusion: look not at the model's list price but at spend per completed task on your own work. The default effort is Medium in Claude Code and the Claude apps, High on the Claude Platform.

Safety and limitations

• Cybersecurity safeguards: Because its cyber capabilities are strong, Sonnet 5.5 is the first Sonnet model to ship with safeguards like Opus 5.5's. Routine software development and bug fixing are unaffected; higher-risk cybersecurity tasks visibly fall back to Sonnet 5. An expanded verification program for cyber defenders was announced.
• Biology: The Sonnet 5 safeguards apply unchanged; some microbiology and virology requests may be flagged in error.
• With thinking off: If you run Sonnet with thinking off, you need to move to the new between_tools setting; details are in Anthropic's migration guide.

What is in the free plan?

According to the video, Sonnet 5.5 is now the free plan's default model. Anthropic's announcement page gives no plan or quota details; check current limits in the Claude app and on the plans page.

Which model when?

• Everyday coding, bug fixes, documents and slides: Sonnet 5.5, enough and fast in most cases.
• Complex, open-ended work needing sustained judgment: Opus 5.5.
• Lowest cost per unit of performance: The video recommends GPT-6.1 Sol; see the GPT-6.1 Sol article.
• Very long output: Google's Gemini 4 Argon can output 1 million tokens; see What is Gemini 4 Argon and the three-model comparison.

If you wonder which agent tool to pick, see the vibe coding tools comparison.

Frequently Asked Questions

When was Claude Sonnet 5.5 released?

On September 28, 2026. It is the second model of the Claude 5.5 family; Haiku 5.5 follows in the coming weeks.

What is Sonnet 5.5's API price?

$2 per million input tokens and $10 per million output tokens; cache reads $0.20 and 5-minute cache writes $2.50. For Opus 5.5 these prices are $4 and $20.

Is Sonnet 5.5 better than Opus 5.5?

On Terminal-Bench 4.0 its score is above Opus's published score; on most benchmarks it is slightly below Opus 5.5. Anthropic says Opus 5.5 stays clearly stronger on complex, open-ended work.

Why can Sonnet 5.5 come out expensive?

According to the Artificial Analysis measurement in the video, because it produces more output tokens per task than rivals. Lowering the effort level cuts cost noticeably.

Is Sonnet 5.5 free?

According to the video, it is the free plan's default model. Usage limits depend on the plan's rules.

Source

• AI Search, Gemini 4, GPT 6.1, Dots, Claude Sonnet 5.5, Ideogram 4.5, Flux 3: AI NEWS (YouTube, October 4, 2026): free plan note, Artificial Analysis index and cost charts.
• Anthropic, Introducing Claude Sonnet 5.5 (September 28, 2026): benchmark table, prices, safeguards, model ID and platforms.