A second model in the 5.5 family
Anthropic released Claude Sonnet 5.5 on September 28, six days after Claude Opus 5.5. It positioned Sonnet for everyday coding, document creation and other well-scoped work. Sonnet 5.5 was available that day in Claude and through the Claude Platform as claude-sonnet-5-5, as well as through AWS, Google Cloud and Microsoft Azure. Anthropic said Claude Haiku 5.5 would follow in the coming weeks; that was a plan, not part of this release. [1][2]
Atlas interpretation: The distinction between Sonnet and Opus here is a deployment choice rather than a simple quality ranking. Anthropic presents Opus 5.5 as the stronger option for complex judgment and Sonnet 5.5 as the quicker option for repeated, bounded tasks. The two models also sit at different per-token prices. [1]
The price stayed put; the workload claim changed
Sonnet 5.5 kept Sonnet 5's list price of $2 per million input tokens and $10 per million output tokens; cache reads were $0.20 per million. Anthropic said its own tests found output more than 30 percent faster and up to 30 percent lower cost per task because the model used fewer tokens. Those are workload claims, not a cut in the price of each token. [1]
Atlas interpretation: Anthropic reported 70.6 percent for Sonnet 5.5 on Terminal-Bench 4.0 against 10.3 percent for Sonnet 5, and described stronger coding and knowledge-work results at selected effort settings. The release page mixes settings across its comparisons and notes a pre-release deployment bug affecting some Artificial Analysis tests. The figures establish what Anthropic reported under those conditions, not a general guarantee for every prompt or a directly comparable independent result. [1]
A Sonnet model with cyber fallbacks
Anthropic said Sonnet 5.5 was the first Sonnet model deployed with cyber safeguards similar to its Opus line. Its announcement says higher-risk cybersecurity requests can visibly fall back to Sonnet 5, while routine software development should continue normally. It retained Sonnet 5's biology safeguards and added measures aimed at attempts to extract model reasoning at scale. [1]
Atlas interpretation: The safety assessment and benchmark comparisons on the launch page are mainly Anthropic-run or Anthropic-reported. TechCrunch independently confirmed the September 28 release and described the cyber fallback, but did not reproduce the benchmark scores or establish savings in customer workloads. The release is a shipped model with measured claims that still need outside validation. [1][2]
Sources
- Introducing Claude Sonnet 5.5
Anthropic · Sep 28, 2026
- Anthropic releases Sonnet 5.5, which it calls a significantly cheaper, faster work partner
TechCrunch · Sep 28, 2026