Anthropic releases Sonnet 5.5, which it calls a significantly cheaper, faster work partner

Lucas Ropek reports Anthropic released Sonnet 5.5, positioning the mid-tier Claude model as significantly faster and cheaper to run than its predecessor from about three months ago. Anthropic claims roughly 30% more speed and a slower token-burn rate.

Benchmarks cited by the company show Sonnet 5.5 beating Opus 5.5 on agentic coding—likely because it can spawn multiple agents without blowing cost limits—while carrying cyber capabilities Anthropic calls comparable to Opus 5. As a result, 5.5 is the first Sonnet subject to the same cyber safeguards applied to Fable and Opus. A new Haiku is planned in coming weeks without a firm date.

We’re here for the workhorse story. Most enterprises will not live on Opus forever; they need a fast mid-tier that can run agent swarms without melting the bill.

Useful AI wins when capability per dollar jumps. A 30% speed claim plus thriftier tokens is how Claude stays in coding and document workflows against OpenAI’s mid-tier refreshes. Putting Opus-class cyber safeguards on Sonnet is the serious part—Anthropic is admitting the mid-tier is now dangerous enough to need the adult railings.

I’m watching independent evals, not only Anthropic’s agentic-coding charts. If cyber parity with Opus is real, defenders and red-teamers should treat Sonnet 5.5 like a frontier-adjacent tool, not a toy.

Haiku’s follow-on will complete the stack refresh. Until then, Sonnet 5.5 is the default I’d trial for multi-agent coding costs—while keeping Opus for the hardest single-thread reasoning.

TechCrunch, Lucas Ropek, September 28, 2026. OpenAI refreshed Sol and Luna the prior week; Meta also announced a new model for smart-glasses features.