Anthropic

Claude Sonnet 5

Anthropic's mid-tier model, now at its best as a workhorse subagent that a stronger model hands focused jobs to.

Quality

Modality

multimodal

Context

1M tokens

Access

closed

Fabian's Take

"When it launched, Sonnet 5 came closer to Opus than any Sonnet generation before, and it's agentic in a way earlier Sonnets weren't: it strings together more tool calls and sticks with harder problems longer. These days I don't pick it by hand at all. It's still a workhorse, though, when a stronger model hands it a focused job as a subagent."

Claude Sonnet 5 is the latest version of Anthropic’s mid-tier model, and it closes most of the gap that used to separate Sonnet from Opus. It’s not just faster and cheaper anymore — it’s meaningfully better at working through multi-step tasks on its own.

How it compares

At launch I ran it against Opus 4.8, and for coding, drafting, and everyday agentic work the two were close enough that Sonnet 5 was the more sensible default. It’s also a clear step up from Sonnet 4.6 in how it handles tools: it uses more of them, in longer sequences, and holds up over longer stretches of autonomous work without drifting off track.

The Opus tier has moved twice since then. Opus 5.5 is the current one, and it’s so capable and efficient that I now use it for almost everything I do by hand. Sonnet’s place has shifted with it. I don’t select it manually anymore, but it still does a lot of work as a subagent: Opus plans, then hands Sonnet a focused job, like searching a codebase, drafting a section, or running a well-defined change, and Sonnet does it quickly at half the per-token price. If you’re building on the API and paying per token for high-volume work, it’s still a strong, cheap default.

What it costs

$2 per million input tokens and $10 per million output. Anthropic launched it at that price as an introductory rate, with a rise to $3 and $15 planned for September, then made the lower price permanent instead. Sonnet 5.5 is announced for the coming weeks.