Anthropic Launches Sonnet 5.5, Claiming 30% Speed Boost

Anthropic has released Sonnet 5.5, a mid-tier AI model it says runs 30 percent faster than its predecessor while burning through tokens at a significantly slower rate.

The speed claim anchors the launch. Sonnet 5, announced about three months ago, sold itself on efficient agentic deployment — the ability to run autonomous agents at a lower cost than competitors offered. Anthropic now positions 5.5 as the faster, cheaper workhorse in the same bracket.

The company describes the new model as an ideal assistant for everyday tasks, including coding and creating office documents. That framing matters commercially: everyday office work, not frontier research, drives volume usage for enterprise customers.

Sonnet sits below Opus in Anthropic's model hierarchy, but rank does not decide usefulness. Anthropic's own benchmarks show Sonnet 5.5 outperforming Opus 5.5 on agentic coding — a result the company attributes to the smaller model's agility. Specifically, Sonnet 5.5 can spawn multiple agents without exceeding cost limits, a constraint that hampers its larger sibling.

The new model also carries cyber capabilities that Anthropic rates as "comparable" to Opus 5, the company's top-tier offering. That parity has a governance consequence. Anthropic says Sonnet 5.5 is the first Sonnet model subject to the same cyber safeguards that apply to Fable and Opus. A mid-tier model crossing a capability threshold normally reserved for frontier models marks a quiet but notable shift in how the lab calibrates risk.

More hardware is on the way. Anthropic plans to release a new version of Haiku, its smallest model, in the coming weeks. The company did not give a firm date.

The release lands amid an unusually crowded stretch for the industry. OpenAI released a batch of new models just last week, including enhanced versions of Sol and Luna, its mid-tier and budget-friendly offerings. Meta announced a new model of its own, which it said would power an upcoming feature tied to its smart glasses.

The pattern across these releases is differentiation by segment rather than raw capability. Labs are refining mid-tier and budget models — the workhorses that handle coding, documents and agent workloads at scale — rather than chasing headline benchmark wins at the frontier. Anthropic's claim that a cheaper model beats Opus on agentic coding fits that trajectory: for customers running fleets of agents, cost per task now matters more than peak intelligence.

The unanswered question is pricing power. If Sonnet 5.5 delivers the claimed speed and token savings, it puts pressure on OpenAI's Sol lineup in the same tier — and sets the stage for Haiku's refresh to compete at the budget end.