Anthropic Launches Claude Sonnet 5.5 With 30%+ Faster Output at the Same API Price

Summary

Anthropic launched Claude Sonnet 5.5 on September 28, positioning it as a faster, more efficient upgrade to Sonnet 5 for coding, well-scoped everyday work, documents, slides, spreadsheets, and agentic workflows.

The sticker price has not changed: $2 per million input tokens, $10 per million output tokens, and $0.20 per million cache-read tokens. The more important claim is efficiency. Anthropic says Sonnet 5.5 generates output 30%+ faster and can cost up to 30% less per task because it often uses fewer tokens and tool calls.

The model is available through the Claude Platform and major cloud platforms, and GitHub says it is already generally available in GitHub Copilot.

Quick Take

• Claude Sonnet 5.5 is available now across Anthropic's platforms and major cloud providers.
• Standard API pricing stays at $2/M input and $10/M output, unchanged from Sonnet 5.
• Anthropic reports 30%+ faster output and up to 30% lower cost per task.
• Sonnet 5.5 has a 1 million-token context window and up to 128,000 output tokens.
• The model adds stronger cyber safeguards and reasoning-extraction protections as its capabilities move closer to Anthropic's higher-end models.

What changed

Anthropic describes Sonnet 5.5 as the second member of the Claude 5.5 family, following Opus 5.5. The company says Opus remains the better choice for difficult, open-ended work requiring sustained judgment, while Sonnet 5.5 is optimized for faster, well-scoped work.

That positioning matters because the upgrade is not primarily a list-price cut. Instead, Anthropic is arguing that users can complete comparable work with fewer tokens, fewer tool calls, and less elapsed time.

GitHub independently says its early testing found Sonnet 5.5 matched Sonnet 5 on coding tasks while using significantly fewer steps, tokens, and tool calls, and finished tasks noticeably faster.

Pricing: same rate card, different economics

The table exposes the main economic story: Sonnet 5.5 costs half as much per standard input and output token as Opus 5.5, but Anthropic says its highest-effort settings can approach Opus-class performance on some evaluations.

That does not mean Sonnet 5.5 is universally the better-value model. Higher effort can consume substantially more tokens, and Anthropic explicitly says Opus 5.5 remains stronger for complex work requiring sustained judgment.

Original-value element 1: sticker price vs. task price

AI API buyers often compare models by cost per million tokens. Sonnet 5.5 is a useful example of why that can be misleading.

If two models have identical token prices but one needs fewer reasoning steps, tool calls, retries, or generated tokens to finish the same job, the effective cost per completed task can be lower even though the rate card is unchanged.

For production agents, cost per successful task is therefore a better metric than token price alone.

Vendor-reported performance

Anthropic reports a large jump over Sonnet 5 on several evaluations. Its headline result is 70.6% on Terminal-Bench 4.0, compared with 10.3% for Sonnet 5. It also reports 55.5% on CursorBench 4.0 and near-Opus results on its GDPval-AA knowledge-work evaluation.

These are vendor-reported benchmark results, not independent AI World Scope testing. Anthropic also cautions that benchmark margins are becoming less reliable as a proxy for real-world differences.

A particularly important footnote is effort level. Sonnet 5.5's best score is not automatically its most economical operating point. Anthropic's own cost-performance charts show that lower effort settings can provide a more attractive trade-off for routine work.

Original-value element 2: Sonnet is becoming a routing tier, not a compromise tier

Historically, choosing a mid-tier model often meant accepting a clear capability penalty in exchange for lower cost.

Sonnet 5.5 narrows that distinction. For organizations building model routers, the practical architecture increasingly looks like:

• Sonnet 5.5 for high-volume, well-scoped coding and knowledge work.
• Opus 5.5 for ambiguous tasks where sustained judgment matters more than unit cost.
• A cheaper fast model for simple classification, extraction, or high-volume low-complexity requests.

That makes model routing more valuable. Instead of choosing one frontier model for every request, teams can reserve expensive reasoning capacity for the minority of tasks that need it.

Safety changes

Anthropic says Sonnet 5.5 improves on or matches Sonnet 5 across most measures in an automated behavioral audit covering roughly 1,850 scenarios.

The company says the model's cybersecurity capabilities have increased enough that Sonnet 5.5 is the first Sonnet model launched with cyber safeguards and fallbacks similar to those used for its more capable models.

Anthropic also added classifiers designed to prevent industrial-scale reasoning extraction. Routine software development is not supposed to be affected, while higher-risk cybersecurity requests can fall back to Sonnet 5.

These are Anthropic's assessments and controls; they should not be read as proof that the model cannot exhibit untested failure modes.

Availability and migration

Claude Sonnet 5.5 is available through the Claude Platform with the API model ID claude-sonnet-5-5, and Anthropic says it is available through Amazon Web Services, Google Cloud, and Microsoft Azure.

GitHub also made Sonnet 5.5 generally available in GitHub Copilot on launch day.

Developers migrating from Sonnet 5 should note that Anthropic documents a change for configurations that run with thinking disabled: they need to use the new between_tools setting before moving to Sonnet 5.5.

Who should care

High-volume agent and coding teams should pay the most attention. If Anthropic's efficiency claims hold on their own workloads, the same list price could translate into meaningfully lower monthly spend and faster completion.

Teams already using Sonnet 5 have a particularly straightforward reason to benchmark the upgrade because the standard API price is unchanged.

Teams doing the hardest open-ended work should not assume Sonnet 5.5 replaces Opus 5.5. Anthropic itself continues to position Opus as the stronger model when sustained judgment is the bottleneck.

AI World Scope take

The most important part of Sonnet 5.5 is not a single benchmark score. It is the combination of unchanged token pricing, higher speed, lower claimed task cost, and near-Opus capability on selected workloads.

That shifts the buying question from “Which model has the lowest token price?” toward “Which model finishes this workload reliably with the least total compute and intervention?”

For teams running agents at scale, that is the metric that ultimately reaches the invoice.