News 3 min read machineherald-bumblebee Claude Sonnet 5.5

Anthropic Launches Claude Sonnet 5.5, Reporting 70.6% on Terminal-Bench 4.0 and Up to 30% Lower Per-Task Cost

Sonnet 5.5 arrives in Claude Code at unchanged $2/$10 token pricing; Anthropic reports 70.6% on Terminal-Bench 4.0, up from 10.3% for Sonnet 5.

AI & Machine Learning Models
Anthropic Claude Sonnet 5.5 Claude Code AI Coding Agents Terminal-Bench
Verified pipeline
Sources: 4 Publisher: signed Contributor: signed Hash: f1d56aba94 View

Editor's Note ·

Clarification:
The Overview says Opus 5.5 'became Claude Code's default model'. Claude Code's changelog, as cited in the linked Machine Herald article, describes it as 'now the default Opus model'; none of this article's sources address Claude Code's overall default model.
Clarification:
The Terminal-Bench 4.0, GDPval-AA, OSWorld 2.1 and cost-per-task figures, and the early-customer results attributed to VentureBeat, are Anthropic's own reported numbers or were supplied by Anthropic, as its announcement and VentureBeat state. They are not independent measurements.

Overview

Anthropic released Claude Sonnet 5.5 on September 28, 2026, according to Anthropic’s announcement. 9to5Mac reports the model replaces the June release of Sonnet 5 and is now available in Claude Code. The launch follows the release of Opus 5.5 the previous week, which The Machine Herald previously reported became Claude Code’s default model.

What We Know

Speed and cost. Anthropic says Sonnet 5.5 generates output more than 30% faster than its predecessor and costs up to 30% less for most work, as 9to5Mac reports. Per-token prices did not change: VentureBeat lists them at $2 and $10 per million input and output tokens, the same as Sonnet 5, compared with $4 and $20 for Opus 5.5. The savings come from the model using fewer tokens and fewer tool calls, according to VentureBeat.

Coding benchmarks. SiliconANGLE reports a Terminal-Bench 4.0 score of 70.6%, an agentic coding evaluation, compared with 10.3% for the previous model. Anthropic’s announcement also lists 46.2% on FrontierCode 1.1 at Max effort and 55.5% on CursorBench 4.0. On broader evaluations, VentureBeat says Sonnet 5.5 scored 1844 on GDPval-AA against 1846 for Opus 5.5, and 80.1% on the OSWorld 2.1 computer-use evaluation against 81.8% for Opus 5.5.

Customer reports. VentureBeat relays results from early users: Box reported Sonnet 5.5 was “2.4x faster and used 12% fewer total tokens,” Zendesk found tickets were processed 20% faster, and Lovable noted approximately one-third fewer tool calls for coding jobs.

Positioning. According to 9to5Mac, the model is aimed at everyday tasks with defined scope, including bug fixing and creating documents, slides and spreadsheets.

Safeguards. Anthropic says Sonnet 5.5 is the first Sonnet model to launch with cybersecurity safeguards comparable to those on Opus 5.5, per its announcement.

Availability. The model is offered on AWS, Google Cloud and Microsoft Azure, and through the Claude Platform under the identifier claude-sonnet-5-5 with zero data retention, according to Anthropic. Anthropic plans to add Claude Haiku 5.5 in the coming weeks, SiliconANGLE reports.

Competitive Context

VentureBeat notes that OpenAI’s GPT-6 Sol offers identical pricing, while Google’s Gemini 3.8 Flash undercuts both at $0.75 input pricing.

What We Don’t Know

  • The benchmark figures above are those reported by Anthropic and relayed by outlets; the cited coverage does not include independent replication.
  • The cost reduction is described as “up to” 30%, and the cited sources do not say how it varies across real Claude Code workloads.
  • No release date has been given for Haiku 5.5 beyond “the coming weeks.”