Claude Sonnet 5.5 launches with unchanged token prices

Claude Sonnet 5.5 launches with unchanged token prices

Ondřej Barták
Ondřej Barták
Entrepreneur and Programmer
29. 9. 2026
3 minutes reading · 14 views
Listen to the article
Audio version of the article
Claude Sonnet 5.5 launches with unchanged token prices

Anthropic has released Claude Sonnet 5.5 across its platforms, keeping the token prices charged for Sonnet 5. The release also introduces a required configuration change for developers upgrading from Sonnet with thinking disabled, alongside different default reasoning levels in Claude apps and the Claude Platform.

Faster output and reported savings per task

Anthropic says Sonnet 5.5 generates output more than 30% faster than Sonnet 5. It also reports that the new model typically uses fewer tokens to complete the same work, with costs per task up to 30% lower than its predecessor in the company’s tests. The reported savings come from reduced token use rather than a cut in the published token rates.

Pricing remains $2 per million input tokens and $10 per million output tokens. Cache reads cost $0.20 per million tokens, also unchanged from Sonnet 5.

Command-line and knowledge-work comparisons

In Anthropic’s published Terminal-Bench 4.0 comparison, Sonnet 5.5 scores 70.6%, against 10.3% for Sonnet 5 and 66.4% for Opus 5.5. The company describes the benchmark as assessing complex, multistep professional tasks performed through a command-line interface. The Opus result uses the Xhigh effort setting, which produced that model’s highest score.

The company’s GDPval-AA v2.1 results show a different comparison with Opus: Sonnet 5.5 scores 1844, above Sonnet 5’s 1449 but below Opus 5.5’s 1846. Artificial Analysis ran GDPval-AA and AA-Briefcase on a prerelease Sonnet 5.5 deployment. Anthropic says that deployment contained a bug that could degrade responses to requests using structured outputs, and that the bug has since been fixed.

Reasoning settings and a coding-test trade-off

The default effort level is Medium in Claude Code and the Claude apps, while the Claude Platform defaults to High. Anthropic describes lower settings as producing faster answers with fewer tokens. At higher settings, the model spends longer reasoning through a task and performs more thorough checks of its work.

More effort did not produce a better result in every reported test. Anthropic says Sonnet 5.5 scored lower in FrontierCode at Max than at Xhigh. That evaluation assesses whether a code change can be accepted without human edits. Its scoring also penalizes changes that go beyond the scope of the assigned task.

According to Anthropic, the model more frequently invoked Claude Code’s code-review skill at Max, dividing review work among multiple subagents. In two cases examined by Cognition, that behavior led either to a timeout or to additional edits outside the assignment, lowering the score.

An internal presentation task

Anthropic also described an internal test centered on preparing a business presentation. The input materials included a publicly traded company’s quarterly earnings documents and transcripts of the related calls. Sonnet 5.5 also received a slide template, with an assignment to produce a ten-slide operating review from those materials. According to the company, two experts judged the first draft ready to send without edits.

Where to access it and what changes on migration

Alongside Anthropic’s own platforms, Sonnet 5.5 is available through Amazon Web Services, Google Cloud and Microsoft Azure. Developers accessing it on the Claude Platform can select the model identifier claude-sonnet-5-5.

Developers who previously ran Sonnet with thinking turned off must switch to the new between_tools setting before moving to Sonnet 5.5. That setting keeps the initial reasoning phase disabled.

The release also expands preserved thinking, which binds the reasoning record to the account that created it. This change particularly affects conversations moved between accounts, including a switch of accounts during an ongoing Claude Code session.

Access limits for higher-risk cybersecurity work

Anthropic says higher-risk cybersecurity tasks will visibly fall back from Sonnet 5.5 to Sonnet 5, while routine work finding and fixing code bugs is to remain available. The company plans to open an expanded Cyber Verification Program offering tiered access to more advanced capabilities.

Advertisement

Content created with help from UpTier.

SEO and GEO on autopilot. UpTier’s multi-agent systems write and optimize content for search engines and AI answers.

Discover UpTier ↗

Category:AI
Did you enjoy this article?
Discover more interesting posts on our blog
Back to blog

Related posts

Google makes Tasks free for Gemini 4 Hydrogen usersGoogle makes Tasks free for Gemini 4 Hydrogen users
Tasks is becoming available for free to hundreds of millions of Gemini 4 Hydrogen users. Google says it can call businesses, book appointments and complete purchases on their behalf.
1 min read
18. 11. 2026
OpenAI introduces GPT-6.2 Astra and Sol for professional and everyday workOpenAI introduces GPT-6.2 Astra and Sol for professional and everyday work
OpenAI has introduced two models: Astra for complex professional tasks and Sol for everyday work. Both are available in ChatGPT Work, Codex and through the API.
3 min read
17. 11. 2026
Tencent releases Hy4 Flash with image support and a context window of up to 1 million tokensTencent releases Hy4 Flash with image support and a context window of up to 1 million tokens
The lighter Hy4 Flash model targets chat, documents and routine code edits. It accepts text and images, supports tool use and, according to Tencent, matched Claude Haiku 5.5 in tests of everyday tasks.
1 min read
16. 11. 2026
Přihlaste se k odběru našeho newsletteru
Zůstaňte informováni o nejnovějších příspěvcích, exkluzivních nabídkách, a aktualizacích.
CodedTrip

Operated by CodedTrip LLC, USA.

YouTube
TikTok