Anthropic ships Sonnet 5.5 with claimed 30% speed boost and slower token burn

Anthropic refreshes its mid-tier model

Anthropic has released Sonnet 5.5, the newest version of the mid-tier model in its lineup, and the pitch behind the update is straightforward: it should get more done in less time for less money. According to TechCrunch, Anthropic claims Sonnet 5.5 is 30 percent faster than its predecessor and consumes tokens at a significantly slower rate, which should translate into lower running costs for everyday work such as coding and producing office documents.

The release comes roughly three months after Sonnet 5 arrived. TechCrunch notes that Sonnet 5's headline selling point was cost-efficient agent deployment — running autonomous agents more cheaply than rival models could. With 5.5, the emphasis shifts toward responsiveness and per-task economics, positioning the model as a practical assistant rather than a frontier showcase.

Where Sonnet sits in the hierarchy

Within Anthropic's model stack, Sonnet is less powerful than Opus, but that ranking does not always hold up in practice because of Sonnet's agility. TechCrunch reports that Anthropic's own benchmarks show Sonnet 5.5 beating Opus 5.5 on agentic coding, a result the company attributes to Sonnet's ability to run several agents at once while keeping total spend in check. For workloads built on orchestrating many parallel agents rather than relying on a single very capable model, the cheaper tier can effectively come out ahead.

Cyber capabilities bring new safeguards

The new model also carries notable cybersecurity capabilities. TechCrunch says Anthropic describes Sonnet 5.5's cyber capabilities as "comparable" to those of Opus 5. That comparison has a regulatory side effect: Anthropic states that Sonnet 5.5 is the first Sonnet model to fall under the same cyber safeguards that already apply to its Fable and Opus models. In effect, capability gains in the mid tier are pulling it into the same safety regime as the company's larger models.

A Haiku refresh and a crowded release cycle

A new version of Haiku, Anthropic's smallest model, is planned for release in the coming weeks, though the company did not commit to a firm date, according to TechCrunch.

The launch also lands in the middle of an unusually busy stretch for the industry. TechCrunch points out that OpenAI rolled out several new models just last week, including upgraded versions of Sol and Luna, its mid-tier and budget-friendly offerings. Meta, meanwhile, announced a model it says will power an upcoming feature tied to its smart glasses.

Why it matters

Mid-tier models are where most production AI work actually runs, so gains in speed and token efficiency feed directly into what developers and enterprises pay at scale. If Sonnet 5.5 really does outperform the larger Opus 5.5 on agentic coding while costing less to run, it reinforces the argument that orchestrating many inexpensive agents — rather than leaning on one expensive model — is the better architecture for coding workflows. The extension of cyber safeguards down to a Sonnet release is equally worth watching: as smaller models inherit frontier-level capabilities, the safety obligations attached to those capabilities appear to be following them down the stack.