Sonnet 5 closes in on Opus 4.8

What Anthropic shipped on 30 June

Anthropic released Claude Sonnet 5 on 30 June 2026, calling it a substantial upgrade to the Sonnet line for agentic work — coding, tool use, multi-step plans and longer autonomous runs. The new model is the default on Free and Pro plans from launch, with availability across Max, Team and Enterprise on the Claude API, Claude Code and the platform. Per the launch post, Sonnet 5 shows notable gains over Sonnet 4.6 on reasoning, tool use, coding and knowledge work, and a lower rate of misaligned behaviour than its predecessor.

There’s a pricing hook. Sonnet 5 ships at an introductory $2 per million input tokens and $10 per million output tokens through 31 August 2026, then $3/$15 standard. Anthropic has also increased rate limits across Chat, Cowork, Claude Code and the Claude Platform to handle the longer agentic runs the new model is designed to take.

Where Sonnet 5 sits on the curve

The interesting shift is where Sonnet 5 lands on the cost-performance curve. Sonnet 4.6 sat well below Opus 4.8 on agentic benchmarks, so the only way to get top accuracy on hard agentic work was to pay Opus prices. Sonnet 5 has closed that step.

Per Anthropic’s framing, Sonnet 5 is close to Opus 4.8 in performance but at lower prices. The two models now cover the same range on the same benchmarks — agentic search and computer use — with Sonnet 5 carrying the lower-cost end and Opus 4.8 still holding the accuracy ceiling. The lever that makes this work is an effort setting on the API; most teams won’t need to touch it, since the default already puts you near the top of the Sonnet 5 range.

Early reactions

Independent commentary on launch day has been broadly aligned with Anthropic’s positioning. X creator Alex Finn, who had been tracking the launch, called Sonnet 5 the new go-to model for the multi-step agentic work he runs in his own stack.

It happened. Claude Sonnet 5 released. Dominates Opus 4.6. Almost as good as Opus 4.8. A fraction of the price. It’s also significantly faster than Opus, while being incredibly agentic. Without a doubt the new go to model for Hermes/OpenClaw

— Alex Finn (@AlexFinn) Jun 30, 2026

An earlier benchmark from Cosmic comparing Sonnet 4.5 and Opus 4.5 found Opus used 19.3% fewer total tokens to build a comparable application — and the team said that token efficiency alone was often enough to narrow the per-token price gap on harder work. Sonnet 5 narrows the capability gap further still.

How it compares on the obvious axes

A few notes on what the launch does and doesn’t change for teams already using Claude:

• Default model swap. Sonnet 5 is the default on Free and Pro; Max, Team and Enterprise users can select it.
• Rate limits raised. Anthropic has lifted Sonnet and Haiku rate limits across all three usage tiers to handle the longer agentic runs Sonnet 5 is designed for.
• A new tokenizer. Sonnet 5 uses an updated tokenizer, similar to the one introduced with Opus 4.7. The same input maps to slightly more tokens than Sonnet 4.6 — Anthropic says the introductory price is set so the move is roughly cost-neutral.
• Cyber safeguards on by default. Cybersecurity work that needs reduced guardrails should still go to Opus 4.8, per Anthropic.

When to pick Sonnet 5 vs Opus 4.8

The decision is now narrower than it was. For a typical UK small team picking a Claude model today:

• Pick Sonnet 5 by default for coding, multi-step tool use, customer support drafting, research, content workflows and most agent jobs. It’s where the cost-performance ratio has shifted sharply, and where most of the new agentic muscle lives.
• Step up to Opus 4.8 when maximum accuracy matters more than price. Long-horizon multi-step work, high-stakes legal or financial analysis, complex code migrations across multiple repos, and any job where a wrong answer costs more than the saving.
• Stay on Sonnet 4.6 (or below) only if you’ve hit a workflow break with the new model and need to roll back.
• Don’t reach for Fable 5 for this work. Fable 5 is the new top tier above Opus 4.8 and costs $10/$50 per million tokens. It’s for jobs that consistently push Opus to its limits, not the everyday agentic layer.

Sonnet 5 is now the workhorse, Opus 4.8 is the specialist, and Fable 5 is the escalator. Most UK small teams will spend most of their budget on Sonnet 5 and reach for Opus 4.8 only when a job demands it.