Claude Sonnet 5.5

3:17 Claude Sonnet 5.5 is the middle model of Anthropic's current Claude lineup, released on Sep 28, 2026 at $2 per million input tokens and $10 per million output tokens, the same rate as Sonnet 5 and half the price of Claude Opus 5.5. Anthropic describes it as "the best combination of speed and intelligence" among its models (Models overview, Anthropic docs, read Nov 6, 2026), and says it is over 30% faster than Sonnet 5 and up to 30% cheaper per task. It is the most capable model on Claude's Free plan.

Quick facts

Sources: Models overview, Pricing and Model deprecations, Anthropic docs, read Nov 6, 2026; Introducing Claude Sonnet 5.5, Anthropic, Sep 28, 2026; Plans and pricing, Claude, read Nov 6, 2026.

What is new in Sonnet 5.5

• Speed and cost. Anthropic reports more than 30% faster output than Sonnet 5 and up to 30% lower cost per task for most work (Introducing Claude Sonnet 5.5).
• Cheaper cache reads. On Oct 7, 2026 Anthropic halved Sonnet 5.5's cache read price from $0.20 to $0.10 per million tokens, which it says cuts about 20% off most agentic tasks. The Sonnet 5.5 post now shows the newer price.
• Effort levels recalibrated. Low, Medium, High, Xhigh and Max no longer produce the same amount of thinking as on Sonnet 5 (Effort, Anthropic docs, read Nov 6, 2026).
• Thinking off becomes between_tools. Sonnet 5.5 rejects disabled; Anthropic asks those who ran Sonnet with thinking off to use between_tools, which skips thinking before the first answer at high effort or below.
• Cyber safeguards. It is the first Sonnet launched with cyber safeguards; higher-risk cyber requests fall back to Sonnet 5.

Anthropic also says it is the first Sonnet to beat Pokémon Red using only screenshots.

Claude Sonnet 5.5 benchmarks

Anthropic's launch table:

Source: Introducing Claude Sonnet 5.5, Anthropic, Sep 28, 2026. Anthropic explains the lower FrontierCode score at max effort by out-of-scope changes and subagent timeouts. Its later Claude Haiku 5.5 post lists Sonnet 5.5 at 1840 on GDPval-AA and 1824 on AA-Briefcase, slightly different from the figures above (Introducing Claude Haiku 5.5, Oct 7, 2026).

The independent Terminal-Bench 4.0 leaderboard does not match Anthropic's 70.6%: it shows Sonnet 5.5 at max effort at 61.82%, about 9 points lower and second behind Opus 5.5 (read Nov 6, 2026). Artificial Analysis gives it 55 at max effort on its Intelligence Index and 42 at high, at $1.64 per task at high. On Arena, Sonnet 5.5 at xhigh is third in WebDev at 1770 but 31st in Text Arena at 1477, below several older Opus models (tables dated Oct 27, 2026).

Sonnet 5.5 vs Opus 5.5

Sources: Models overview, Anthropic docs; Terminal-Bench leaderboard; Models leaderboard, Artificial Analysis; all read Nov 6, 2026; Text Arena, Oct 27, 2026.

1:02 Choose Sonnet 5.5 when cost and speed matter and the task is ordinary coding, writing or agent work: in Anthropic's tests it lands within a few points of Opus 5.5 on most rows at half the price. Choose Opus 5.5 for long, hard tasks where independent tests still show a lead, or where its higher text-chat ranking matters. Anthropic's own pricing page calls Opus 5.5 the "daily driver" for agentic coding and Sonnet 5.5 a "high-performance model for coding and agents" (Plans and pricing, read Nov 6, 2026).

Where Claude Sonnet 5.5 is available

Sonnet 5.5 is on every Claude plan, Free included (Plans and pricing, read Nov 6, 2026); Anthropic's pages read for this article do not say which model the apps open with by default. In Claude Code it runs at Medium effort by default and is chosen with /model sonnet. It is on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS, with zero data retention available (Introducing Claude Sonnet 5.5).

Safety

Anthropic tested Sonnet 5.5 on about 1,850 automated behavioral audit scenarios and calls it the least likely of its models to probe the limits of its sandbox. It is also the first Sonnet with classifiers that block attempts to extract its reasoning (Introducing Claude Sonnet 5.5).

What reviewers found

24:44 In a Sep 28, 2026 video, Nate Herk of Nate Herk | AI Automation gave Sonnet 5.5 and Opus 5.5 the same seven tasks at high effort, billed at API prices. Opus gave the better result in most rounds, above all from vague, open-ended prompts such as a 90-day growth plan, but he credited the cheaper model when he judged a second Sonnet prompt could close the gap, so Sonnet won four rounds to three. Across the seven runs Opus took about 46 minutes longer and cost about $16 more, well short of double, and his rule of thumb was Sonnet for tasks with a clear definition of done and Opus for creative, open-ended work.

7:32 Theo Browne of Theo - t3.gg, in a Sep 29, 2026 video, faulted the launch price: Sonnet 5.5's $0.20 cache reads had not been cut the way Fable 5.1's and Opus 5.5's were, and since cache reads make up most of an agent's bill, he found Sonnet costing as much as Opus or more on like-for-like code work. He also noted 271,920 tokens per task on CursorBench, more than Opus's 218K, and in his own game-building test Sonnet took 43 minutes to Opus's 36. Anthropic halved the cache read price on Oct 7, 2026, after his video.

22:58 Where Theo did rate it was as a helper. In his test auditing a very large codebase, Sonnet 5.5 cost about half as much as Opus 5.5, scored slightly higher with his automated judges and finished in about five minutes. His advice was not to pick Sonnet yourself but to let Opus or Fable call it for research subagents; he was not impressed with its front-end design.

Frequently asked questions

How much does Claude Sonnet 5.5 cost? Claude Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens on the Claude API, and $1 and $5 through the Batch API. It is free to use, within limits, on Claude's Free plan.

Is Claude Sonnet 5.5 free? Yes. Claude Sonnet 5.5 is included on the Free plan, with limits every five hours and every week; paid plans raise those limits.

Should I use Claude Sonnet 5.5 or Opus 5.5? Claude Sonnet 5.5 costs half as much and is faster; Claude Opus 5.5 scores a few points higher on most independent tests. Start with Sonnet 5.5 for everyday work and switch to Opus 5.5 when a task stalls.

Can I turn off thinking on Claude Sonnet 5.5? Partly. On the API, Claude Sonnet 5.5 accepts between_tools, which skips thinking before its first answer at high effort or below, but not a full disabled setting.

When was Claude Sonnet 5.5 released? Anthropic released Claude Sonnet 5.5 on Sep 28, 2026, six days after Claude Opus 5.5.

Sources

• Introducing Claude Sonnet 5.5, Anthropic, Sep 28, 2026
• Introducing Claude Haiku 5.5, Anthropic, Oct 7, 2026
• Models overview, Anthropic docs, read Nov 6, 2026
• Pricing, Anthropic docs, read Nov 6, 2026
• Effort, Anthropic docs, read Nov 6, 2026
• Model deprecations, Anthropic docs, read Nov 6, 2026
• Plans and pricing, Claude, read Nov 6, 2026
• Terminal-Bench leaderboard, tbench.ai, read Nov 6, 2026
• Models leaderboard, Artificial Analysis, read Nov 6, 2026
• Text Arena and WebDev Arena, Arena, Oct 27, 2026
• I Tested Sonnet 5.5 vs Opus 5.5. What You Need to Know., Nate Herk | AI Automation, Sep 28, 2026 (video)
• OpenAI should be scared of this one, Theo - t3.gg, Sep 29, 2026 (video)