# Anthropic's Claude Sonnet 5.5 Brings Opus-Level Agentic Coding at Lower Cost

> Anthropic's Claude Sonnet 5.5 pairs near-flagship agentic coding scores with $2-per-million-token pricing, resetting the mid-tier benchmark as the Claude 5.5 family expands.

*Published 2026-09-29 · By Marcus Vance*

Claude Sonnet 5.5 is Anthropic's second model in the Claude 5.5 family, released September 28, 2026, with a 1,000,000-token context window and pricing of $2 per million input tokens and $10 per million output tokens.

Anthropic released Claude Sonnet 5.5 on September 28, 2026, adding a second model to the Claude 5.5 family as the company presses forward with its commercial expansion. Sonnet 5.5 is built for high-volume coding and knowledge work, and Anthropic says it is a clear upgrade over Claude Sonnet 5, running 30%+ faster and costing up to 30% less for most work. The launch lands after Opus 5.5 and gives enterprises a lower-cost tier that, on at least one benchmark, out-scores the flagship.

The release extends Anthropic's reach across the major enterprise clouds. Claude Sonnet 5.5 is available through the Anthropic API, Amazon Bedrock, Google Cloud, and Microsoft Foundry, a distribution footprint that matches the broadest releases in the industry. For teams already standardized on one of those platforms, the upgrade path is a model ID change rather than a migration.

What does Claude Sonnet 5.5 change for developers?

The most immediate change is price-to-performance. Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens, with batch API requests at 50% off, or $1 and $5 per million tokens. That pricing puts a model with a 1 million-token context window into a range where teams can run large-scale agentic workloads without watching API costs balloon.

Speed is the second change. Anthropic says Sonnet 5.5 generates outputs 30%+ faster than Sonnet 5 and typically costs up to 30% less per task. For interactive coding assistants, latency is a feature; for long-running agents, cost per task is the metric that determines whether a workflow is viable.

The third change is capability. Sonnet 5.5 scored 70.6% on Terminal-Bench 4.0, an agentic coding benchmark that measures a model's ability to complete real coding tasks. That compares with 66.4% for Opus 5.5 and 10.3% for Claude Sonnet 5.

On knowledge work, Sonnet 5.5 reaches 1844 Elo on GDPval-AA, Anthropic's benchmark for realistic office tasks. The combination of a strong agentic coding score and a strong knowledge work score suggests Anthropic is targeting the two largest categories of enterprise AI usage with a single mid-tier model.

How does Claude Sonnet 5.5 compare with Claude Opus 5.5 and Claude Sonnet 5?

The Terminal-Bench 4.0 result is the headline comparison. A Sonnet-tier model beating the flagship on an agentic benchmark is unusual, and the margin, 70.6% to 66.4%, is meaningful. The 10.3% score for Sonnet 5 shows how much the new model covers in one generation.

Sonnet 5.5 also brings writing and reasoning upgrades that Anthropic says carry over from Opus 5.5. The launch materials describe the model as more fun to work with, in part because it produces more natural language. For coding assistants, that matters: clearer explanations, commit messages, and code comments reduce the time developers spend interpreting model output.

The model ID is claude-sonnet-5-5, and the same weights are served across the Anthropic API and the major clouds. That consistency matters for teams that develop on one platform and deploy on another.

What are Claude Sonnet 5.5's technical specifications?

Claude Sonnet 5.5 carries a 1,000,000-token context window and a maximum output of 128,000 tokens. The context length matches the top of the Claude 5.5 family and allows the model to ingest large codebases, multi-file repositories, and long transcripts in a single pass.

The 128K output limit is worth attention. Agentic coding tasks often require long generated responses, including multiple file edits, test cases, and explanations. A model that can write long outputs without truncation is easier to use for autonomous tasks that previously required stitching together multiple generations.

| Attribute | Claude Sonnet 5.5 |
| --- | --- |
| Release date | September 28, 2026 |
| Model ID | claude-sonnet-5-5 |
| Context window | 1,000,000 tokens |
| Maximum output | 128,000 tokens |
| Input price | $2 per million tokens |
| Output price | $10 per million tokens |
| Batch API price | $1 input / $5 output per million tokens |
| Terminal-Bench 4.0 | 70.6% |
| GDPval-AA | 1844 Elo |

The 1M context window also reduces reliance on retrieval pipelines. Teams can attach a full repository or a long set of documentation to a prompt and let the model find what it needs, rather than building chunking and embedding infrastructure. That lowers the barrier to entry for teams that want to try agentic coding.

What does Terminal-Bench 4.0 measure?

Terminal-Bench 4.0 evaluates models on terminal-based coding tasks that require the model to navigate a file system, execute commands, and iterate on errors. It is part of a newer class of benchmarks that test execution and repair rather than static code generation. The leap from 10.3% for Sonnet 5 to 70.6% for Sonnet 5.5 indicates a fundamental improvement in how the model handles autonomous work, not just a better code-writing model.

Benchmarks reported by model developers should be read with the caveat that the vendor controls the evaluation setup. Still, Terminal-Bench 4.0 is an external benchmark, and Anthropic's reported score gives buyers a common reference point against the previous Sonnet generation and against Opus 5.5.

Why does the GDPval-AA score matter for knowledge work?

GDPval-AA is designed to measure performance on realistic knowledge work, and Sonnet 5.5's 1844 Elo places it in range of models that cost more to operate. For teams that use models for document analysis, data extraction, and report generation, the combination of a high Elo and $2 input pricing is the key commercial signal.

Knowledge work benchmarks matter less for raw coding teams and more for enterprises that route a mix of tasks through one model. A single Sonnet 5.5 deployment can handle code generation, documentation, data cleanup, and email drafting, which simplifies model governance and reduces the number of vendor integrations.

How much does Claude Sonnet 5.5 cost and where is it available?

Pricing is $2 per million input tokens and $10 per million output tokens, with batch pricing at $1 and $5. The batch discount is 50% off, per Anthropic's platform documentation. For workloads that can tolerate asynchronous processing, batch pricing is likely to be the default.

- Anthropic API, using model ID claude-sonnet-5-5
- Amazon Bedrock, via Amazon Web Services
- Google Cloud
- Microsoft Foundry

Amazon Web Services announced same-day availability of Claude Sonnet 5.5 on Amazon Bedrock and Claude Platform on AWS. AWS describes the model as 'a smarter, more efficient Sonnet model suited for focused coding and knowledge work with lower cost per task for most work at faster speed.'

Google Cloud and Microsoft Foundry also offer the model. For enterprises, that means procurement, security review, and data governance can stay within their existing cloud relationships.

Batch pricing makes Sonnet 5.5 attractive for offline pipelines, including document classification, data extraction, log analysis, and overnight code review. At $1 per million input tokens, the economics of large-scale processing shift materially.

How do enterprise users describe Claude Sonnet 5.5 in practice?

Epic's early testing suggests the model holds up on demanding engineering work. Daniel Vogel, Chief Operating Officer at Epic, said in Anthropic's launch materials that Sonnet 5.5 'cleared the same quality bar you’d expect from a higher-tier model, holding up on a system design audit and a data flow review.'

Vogel also said the model 'managed tens of thousands of lines of code for gameplay system architecture, kept responses snappy, handled multi-hour tasks, and delivered with less prescriptive prompting.' The multi-hour task point matters because agentic systems are judged on endurance as much as raw accuracy.

Tyler Nishida, a designer quoted in Anthropic's launch post, said Sonnet 5.5 carries some of Opus 5.5's writing upgrades, which he said makes it more enjoyable to work with. The design community's reaction points to a softer quality that still has practical value: better explanations and comments reduce debugging time.

What security changes come with Claude Sonnet 5.5?

Sonnet 5.5 is the first Sonnet model with cyber safeguards and fallbacks comparable to Opus models, according to Anthropic. Mid-tier models historically received lighter safety configurations, and cyber capabilities are a sensitive area for frontier labs, so aligning the tier with Opus-level protections is a notable policy shift.

For enterprise buyers, consistent safety behavior across model tiers simplifies compliance. Companies in regulated industries often need the same guardrails on the models that process customer data, and Sonnet 5.5 reduces the gap between what the flagship offers and what the mid-tier offers.

Anthropic did not detail the specific controls in its launch announcement. The classification of the model's safety features as comparable to Opus suggests a unified safety posture across the Claude 5.5 family.

What does Claude Sonnet 5.5 mean for the broader model market?

The release intensifies pricing pressure across the frontier-model market. A model that beats the previous flagship on an agentic coding benchmark while charging $2 per million input tokens resets expectations for what a mid-tier model should cost.

It also validates the trend toward tiered model families. Anthropic now offers Sonnet 5.5 for high-volume work and Opus 5.5 for the most demanding tasks, while older models like Sonnet 5 remain available. That tiering lets Anthropic capture more of the market without discounting the flagship.

For cloud providers, the multi-platform launch makes Sonnet 5.5 a standard option across the three largest enterprise clouds. AWS's announcement emphasizes lower cost per task, signaling that the model's commercial pitch is about efficiency as much as raw capability.

How should teams evaluate Claude Sonnet 5.5 before adopting it?

Anthropic's reported benchmarks are a starting point, not a guarantee. Teams with existing agentic workflows should run Sonnet 5.5 against their own task sets, using the batch API for low-cost evaluation. The model ID change is trivial, so the marginal cost of testing is low.

Enterprises with strict data governance requirements will need to confirm that the deployment path on their chosen cloud meets their residency and logging policies. Because Sonnet 5.5 is available on multiple clouds, teams can compare those properties side by side.

What role does Sonnet 5.5 play in Anthropic's IPO preparation?

Anthropic has not set a date for a public listing, but a broad, low-priced model release is the kind of product move that expands usage volume ahead of a filing. Sonnet 5.5 gives Anthropic a defensible answer to the question of where growth comes from after the flagship tier saturates.

The multi-cloud availability also supports revenue diversification. By placing the same model on Amazon Bedrock, Google Cloud, and Microsoft Foundry, Anthropic avoids tying its growth to a single cloud relationship and makes it easier for enterprises to spend on Claude through existing contracts.

For investors, the key metrics will be usage growth and cost per task. Sonnet 5.5's pricing is designed to maximize adoption; the question is whether that adoption converts into durable revenue at scale.

What comes next for the Claude 5.5 family?

The Claude 5.5 family now has two members, and Anthropic's release cadence suggests the company is moving faster than its previous annual upgrade cycle. The Sonnet 5.5 launch, roughly a month after Opus 5.5, points to a family-based rollout in which tiers arrive separately.

Anthropic's pricing also leaves room for future moves. The $2 input price for a 1 million-token context model is aggressive, and further efficiency gains could allow lower prices or expanded context windows in future releases. For now, Sonnet 5.5 gives developers a reason to re-test agentic coding workloads against a mid-tier price point.

The next milestones for the company are likely to be enterprise adoption, safety tooling around the Claude 5.5 family, and the IPO process. Sonnet 5.5 gives Anthropic a product that can carry usage volume while the company builds out the rest of the family.

## Sources

1. [Claude Sonnet 5.5 is a clear upgrade over Claude Sonnet 5, runs 30%+ faster, and costs up to 30% less for most work.](https://www.anthropic.com/claude-sonnet-5-5)
2. [Released September 28, 2026. Model ID: claude-sonnet-5-5. Context window: 1M tokens. Input pricing: $2 / MTok. Output pricing: $10 / MTok. Batch API requests are 50% off.](https://platform.claude.com/docs/en/models/sonnet-5-5/overview.md)
3. [Claude Sonnet 5.5 is available on Amazon Bedrock and Claude Platform on AWS, and is suited for focused coding and knowledge work with lower cost per task at faster speed.](https://aws.amazon.com/blogs/machine-learning/introducing-claude-sonnet-5-5-on-aws/)
4. [Anthropic launched Claude Sonnet 5.5 on September 28, 2026, with 1,000,000-token context window, $2.00 input / $10.00 output per million tokens (batch pricing $1/$5), available via Anthropic API as claude-sonnet-5-5.](https://benchlm.ai/model-updates)

---
Source: https://aiintelreport.com/frontier-models/anthropic-claude-sonnet-5-5-agentic-coding
Index: https://aiintelreport.com/llms.txt · Full text: https://aiintelreport.com/llms-full.txt