Anthropic’s Claude Sonnet 5 pushes agentic power down in price

TL;DR:

• Anthropic has released Claude Sonnet 5, its most agentic mid-tier model, with performance approaching the flagship Opus 4.8 at lower cost.
• It launches at introductory pricing of $2 per million input tokens and $10 per million output tokens until 31 August, and is now the default model on Free and Pro plans.
• Anthropic says the model refuses malicious requests more reliably than its predecessor and ships with cyber safeguards enabled by default.

Anthropic has positioned Claude Sonnet 5 as the model that narrows the gap between its cheaper Sonnet line and its expensive Opus flagship. The company says it can plan, use tools such as browsers and terminals, and run autonomously at a level that until recently required larger, costlier models.

Cheaper autonomy, not just cheaper tokens

The pitch is less about raw benchmark scores than about cost-controlled autonomy. Anthropic says users can dial the “effort” level up or down to trade cost against performance, with higher-effort runs matching Opus 4.8 on some tasks. Early testers describe the model completing multi-step jobs — updating records, then sending a launch announcement — that previous Sonnet versions would abandon halfway.

That matters for the wider squeeze on AI budgets. As Resultsense reported this week, rising inference bills are pushing businesses towards cheaper models, and a capable mid-tier option undercuts the case for always reaching for a flagship. A tokeniser change means the same text can map to slightly more tokens, so Anthropic has set introductory pricing to make the switch roughly cost-neutral.

For UK teams building agents, the appeal is a model that finishes work without constant supervision at a price that survives contact with a finance director. Anthropic also reports lower rates of hallucination and sycophancy than Sonnet 4.6, and says the model resists prompt-injection hijacks more effectively.

Looking forward

Anthropic launched Sonnet 5 with cyber safeguards on by default — the same controls used in its Opus models — reflecting a cautious posture as capability spreads down its range. The commercial test now is whether a “good enough” agentic model at a third of flagship prices reshapes how firms budget for AI work over the rest of the year.