Claude Sonnet 5.5 Draws Close to Opus Performance, With a Heavier Token Bill

Claude Sonnet 5.5 came out from Anthropic on September 28 at the same list price as Sonnet 5. The company says it runs more than 30% faster and cuts per-task costs by up to 30%.

A separate benchmarking outfit came to a different view on expenses when it stressed the model to its limits.

Introducing Claude Sonnet 5.5, the second model in the Claude 5.5 family.

It’s a clear upgrade over Sonnet 5, runs more than 30% faster, and costs up to 30% less for most work.

— Claude (@claudeai) September 28, 2026

Sonnet 5.5 Debuts Days After Opus 5.5 in Claude Lineup

The Claude 5.5 family now has its second entry in Sonnet 5.5. Opus 5.5 debuted on September 22, and OpenAI rolled out GPT-6 Sol and Luna that same day.

Pricing carries over from Sonnet 5: $2 per million input tokens and $10 per million output tokens. Anthropic logged a 70.6% result on Terminal-Bench 4.0, an agentic coding benchmark. Sonnet 5 came in at 10.3%, while Opus 5.5 posted 66.4%.

“Where Opus 5.5 is built for complex work requiring careful judgment, Sonnet 5.5 is strongest at well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets,” the Anthropic team wrote.

Sonnet 5.5, in Anthropic’s telling, does not push the capability frontier outward. As a result, the alignment assessment concentrated on hazards like deceiving users and supporting high-stakes abuse.

The release also includes cyber protections and anti-distillation classifiers designed to block efforts to extract the model’s reasoning for training competing systems. Claude Haiku 5.5 arrives in the coming weeks.

OpenAI, in the meantime, put GPT-6.1 Astra on hold for safety reasons. CNBC verified on Monday that the model did not meet the company’s standards.

Artificial Analysis Spots Heavier Token Consumption at Maximum Effort

Benchmarker Artificial Analysis, which placed Grok 4.7 fourth, awarded Sonnet 5.5 a 56 on the Intelligence Index. That is good for second place overall, 2 points shy of Opus 5.5 at max effort.

Artificial Analysis logged 64% for Sonnet 5.5 on Terminal-Bench 4.0, compared with 60% for Opus 5.5 and GPT-6 Astra. On GDPval-AA and similar knowledge-work tasks, the firm saw Sonnet 5.5 roughly on par with Opus 5.5.

According to Artificial Analysis, Sonnet 5.5 consumed considerably more tokens to get there.

“At max effort, where it reaches performance nearing that of Opus 5.5, Claude Sonnet 5.5 used ~193k Output Tokens per Intelligence Index Task,” the firm wrote.

That consumption sits roughly 60% higher than Opus 5.5 and Sonnet 5 at maximum effort, and about 7 times GPT-6 Astra’s max-effort count.

Anthropic’s early-access customers, by contrast, described the reverse pattern relative to Sonnet 5, although those workloads differed. Balyasny Asset Management measured token use far lower on its finance workloads.

“On our private suite of 2,441 finance tasks covering Q&A, extraction, analysis, and forecasting, Claude Sonnet 5.5 scored ahead of Sonnet 5 and used about 121k tokens per answer, where Sonnet 5 used 497k,” according to Joe Poirier, Senior AI Engineer at Balyasny Asset Management.

Artificial Analysis tested a pre-release version with a structured-output bug that Anthropic has since patched. The firm intends to redo the affected assessments shortly.