Anthropic Ships Claude Sonnet 5 — Near-Flagship Performance at a Fraction of the Price

by Geth Author

On June 30th, 2026, Anthropic launched Claude Sonnet 5, making it the new default model for all Free and Pro plan users on Claude.ai. It is the most capable mid-tier model the company has shipped — and arguably the most significant efficiency leap in Anthropic's lineup to date.

Near-Flagship Performance at Mid-Tier Prices

Sonnet 5's headline claim is that its performance sits close to Opus 4.8, Anthropic's top model, but at substantially lower cost. Benchmarks show marked improvements over Sonnet 4.6 on reasoning, software coding, tool use, and knowledge-intensive tasks. It's the model Anthropic is betting enterprises will reach for when they want Opus-level quality without the Opus-level invoice.

For developers, the introductory API pricing runs $2 per million input tokens and $10 per million output tokens through August 31, 2026, after which it moves to standard pricing at $3 and $15 respectively. That positions Sonnet 5 as the obvious first choice for high-volume agentic workloads — tasks that were previously priced out of reach using Opus.

Built for Agents First

What distinguishes Sonnet 5 from its predecessor is not just raw benchmark scores — it's how the model is designed to operate autonomously. According to Anthropic, Sonnet 5 is built to be the most agentic Sonnet model yet — capable of making multi-step plans, using tools like browsers and terminals, and running long autonomous tasks at a level that previously required larger, more expensive models.

This is the real story. As AI pipelines mature, the cost of the underlying model increasingly determines whether an agentic product is viable at scale. A model that can plan, execute, and self-correct across dozens of steps, at $2 per million input tokens, changes the math for a lot of developers.

Safety by Default

Safety assessments show that Sonnet 5 has an overall lower rate of undesirable behaviors than Sonnet 4.6 and is generally safer to deploy in agentic contexts. Cybersecurity safeguards are enabled by default, with real-time detection and blocking of dangerous activity — the same protections available in Opus 4.7 and 4.8. Evaluations also show the model has significantly reduced capability to assist with offensive cybersecurity tasks compared to current Opus models.

For enterprises deploying agents that browse the web, run code, or interact with external APIs, built-in safety rails reduce the compliance friction that tends to slow adoption.

What It Means in Practice

Claude Sonnet 5 is available now across all Claude plans and via the API using the model ID claude-sonnet-5. It is also available to Max, Team, and Enterprise users alongside the existing Opus 4.8 option. If you have been holding back agentic projects because Opus pricing made them impractical at scale, the window to revisit that calculation just opened.