Claude Sonnet 5.5 Released! What You Need to Know
Claude Sonnet 5.5 is much faster and cheaper than Sonnet 5, while delivering performance close to Opus 5.5.
Contents 8
Claude Sonnet 5.5 is Anthropic's new mid-tier AI model, announced on September 26, 2026. It runs both faster and cheaper per task than Sonnet 5. Its significance for businesses lies in delivering near-Opus-level performance at a much more accessible price.
In short:
- Agentic coding success on Terminal-Bench 4.0 jumped from 10.3% to 70.6%.
- Same price as Sonnet 5 ($2/$10 per million tokens), but up to 30% cheaper per-task cost.
- Came within just 2 points of Opus 5.5 on GDPval-AA v2.1.
- Immediately available on AWS, Google Cloud, and Microsoft Azure.
What is Claude Sonnet 5.5?
This model is the newest member of Anthropic's Sonnet family. It sits in the company's mid-tier lineup, positioned as lighter than Opus but stronger than Haiku. Its model ID is listed on platforms as claude-sonnet-5-5. According to Anthropic's announcement, the model makes a major leap specifically in agentic tasks.
Its most notable feature is support for a 1 million token context window. This means it can process long codebases or extensive documents in a single pass. Compared to the previous Sonnet 4.5 release, this represents a substantial architectural overhaul.
How much better is Sonnet 5.5 than Sonnet 5?
The gap is especially dramatic in agentic coding tasks. On the Terminal-Bench 4.0 test, Sonnet 5 managed only 10.3% success, while Sonnet 5.5 reached 70.6%. That's nearly a sevenfold improvement on real-world terminal tasks.
There's a similar leap on the speed side. According to Unite.AI's report, the new model runs more than 30% faster than its predecessor. At the same time, it cuts per-task cost by up to 30%, since it consumes fewer tokens.
This chart makes clear just how large the gap is between the two generations. In complex tasks like agentic coding, the model is now a far more reliable choice.
How has pricing changed?
The price list itself hasn't changed, but the effective cost has dropped. Anthropic priced Sonnet 5.5 at $2 per million input tokens and $10 per million output tokens — exactly the same list price as Sonnet 5.
Cache reads are kept at $0.20 per million tokens. The real cost advantage comes from completing the same task with fewer tokens. Teams using this model can get more done within the same budget.
This approach resembles the strategy also seen in our price-and-performance analysis of Claude Opus 5.5. Anthropic is keeping list prices fixed and competing on efficiency instead.
How close is it to Opus 5.5?
The results show the new model has closed the gap with Opus to an astonishing degree. On the GDPval-AA v2.1 test, Sonnet 5.5 scored 1844 points, while Opus 5.5 leads by just two points with 1846. In practice, that difference is barely noticeable.
A similar picture emerges on the AA-Briefcase v1.1 test. The model scores 1811, close to Opus 5.5's 1822. By contrast, the older Sonnet 5 only managed 1359 points, so the leap is quite large.
On the Humanity's Last Exam test, this model achieved 64.5% success with tool use. However, for complex, long-horizon research tasks, Opus 5.5 still holds the edge. This distinction matters when deciding which model to choose for which purpose.
| Benchmark | Sonnet 5.5 | Opus 5.5 | Sonnet 5 |
|---|---|---|---|
| GDPval-AA v2.1 | 1844 | 1846 | - |
| AA-Briefcase v1.1 | 1811 | 1822 | 1359 |
| Terminal-Bench 4.0 | 70.6% | - | 10.3% |
Should I be worried about cybersecurity?
Anthropic has taken a proactive step on this front. This is the first Sonnet model to launch with safety classifiers that block reasoning extraction attempts. This makes it harder to misuse the model's internal thought process.
There's also improvement on the cybersecurity performance side. The model achieves an 80% capture rate on the ExploitBench test, approaching Opus 5 level results. However, for high-risk tasks, the system automatically falls back to Sonnet 5.
This protective layer is an important trust factor for enterprise users. For teams working with sensitive data, it means an added layer of security.
Is it available right now?
Yes, the model has been available since the day of the announcement. It can be accessed via AWS, Google Cloud, and Microsoft Azure. According to The Decoder's report, this broad rollout makes enterprise adoption easier.
Developers can migrate their existing Sonnet 5 integrations to the new model with minimal changes. Since the API structure remains largely the same, the transition process moves quickly. We touched on similar expectations earlier in our piece about the leaks.
Frequently asked questions
Which model family does Claude Sonnet 5.5 belong to?
Claude Sonnet 5.5 is the latest member of Anthropic's mid-tier Sonnet family. It's positioned as more affordable than Opus and more capable than Haiku.
Do I need a separate API key for Sonnet 5.5?
No, it can be used through the existing Anthropic API infrastructure simply by changing the model ID. Access works the same way via AWS, Google Cloud, and Azure.
Can Sonnet 5.5 replace Opus 5.5?
For standard tasks, often yes, since the performance gap is very small. However, for long-horizon, complex research projects, Opus 5.5 is still the preferred choice.
Did the price actually drop, or did it stay the same?
The list price stayed exactly the same as Sonnet 5. But since the model consumes fewer tokens, the actual per-task cost drops by up to 30%.
Sources
This new model redefines the balance between performance and cost in the mid-tier segment. At EngerekTech, we closely follow new releases to help you choose the right AI infrastructure for your projects.
