Anthropic Launches Claude Sonnet 5.5, Claims 30% Faster Performance at Lower Cost Than Predecessor

Anthropic has released Claude Sonnet 5.5, the newest version of its mid-tier model, positioning it as a faster and cheaper alternative to its predecessor for everyday work tasks including coding and document creation.

Sonnet 5, released roughly three months ago, was marketed around efficient agentic deployment — the ability to run autonomous agents at lower cost than rival models. With 5.5, Anthropic is emphasizing raw speed. The company claims Sonnet 5.5 runs 30 percent faster than Sonnet 5 and burns through tokens at a significantly slower rate, though Anthropic has not disclosed exact pricing figures or a specific token-burn percentage.

In Anthropic's model lineup, Sonnet sits below Opus in raw capability but is often more practical due to its speed and cost profile. According to Anthropic's own benchmarks, Sonnet 5.5 now outperforms Opus 5.5 specifically on agentic coding tasks — a result the company attributes to Sonnet's ability to spawn multiple agents in parallel without breaching cost limits that would constrain the larger Opus model.

Anthropic also disclosed that Sonnet 5.5 carries cyber capabilities it describes as "comparable" to Opus 5. As a result, this is the first Sonnet-tier model that will be subject to the same cyber safeguards Anthropic applies to its Opus and Fable models — a signal that the company views Sonnet 5.5's offensive or dual-use technical capabilities as having crossed a threshold requiring additional oversight.

Anthropic did not release specific figures for context window size, exact input/output pricing per million tokens, or standardized benchmark scores (e.g., MMLU, HumanEval) alongside this launch. The company has not disclosed a training data cutoff date for Sonnet 5.5.

Anthropic also confirmed plans to release an updated version of Haiku, its smallest and cheapest model, in the coming weeks, though it did not commit to a specific release date.

The release lands amid an active stretch of model launches across the industry. Last week, OpenAI shipped updated versions of Sol and Luna, its mid-tier and budget-focused models. Meta also announced a new model intended to power an upcoming feature tied to its smart glasses line.

What this means

Anthropic's messaging with Sonnet 5.5 marks a shift from cost-efficiency (Sonnet 5's pitch) to speed as the primary differentiator — a sign that the agentic coding market, where multiple labs are racing to make mid-tier models viable for parallel agent workflows, has become the main battleground rather than flagship model capability.

The decision to extend Opus-level cyber safeguards to a mid-tier model is notable. It suggests either that Sonnet 5.5's underlying capabilities have advanced substantially, or that Anthropic is tightening its risk thresholds industry-wide as models proliferate. Either reading points to growing scrutiny of dual-use capabilities in models positioned as everyday productivity tools rather than frontier research systems.

The absence of disclosed pricing, context window size, and standardized benchmark numbers in this announcement is a gap worth flagging — claims of being "significantly cheaper" and "30 percent faster" are not independently verifiable without published rate cards or third-party benchmark results. Buyers evaluating a switch from Sonnet 5 should wait for concrete pricing and context-length specs before making procurement decisions.