What launched
On June 30, 2026, Anthropic released Claude Sonnet 5, which it calls "the most agentic Sonnet model yet": it plans, uses tools like browsers and terminals and works autonomously on tasks that, just months earlier, required larger and more expensive models.
According to Anthropic, its performance is close to Claude Opus 4.8 at a lower price, and a clear improvement over Sonnet 4.6 in reasoning, tool use, coding and knowledge work.
Availability and pricing
- It became the default model on Claude's Free and Pro plans and is available on Max, Team and Enterprise.
- In the API, the identifier is claude-sonnet-5.
- Launch pricing: $2 per million input tokens and $10 per million output tokens. On August 10, Anthropic made this pricing permanent, cancelling the planned return to $3/$15 on September 1.
- Note: it uses a new tokenizer, so the same text may map to 1.0–1.35× more tokens, eating into some of the savings.
- Anthropic raised usage limits across Chat, Cowork, Claude Code and the Claude Platform.
Effort levels
Sonnet 5 runs at effort levels (up to "xhigh"). At medium effort it offers the best cost efficiency; at high effort it can match Opus 4.8 on some tasks, according to the BrowseComp and OSWorld-Verified charts Anthropic published.
Safety and cybersecurity
- In internal evaluations it showed fewer undesirable behaviors, less hallucination and less sycophancy than Sonnet 4.6, and better resistance to prompt injection.
- Anthropic says it did not deliberately train the model on cybersecurity. In the Firefox 147 exploit-development test it never produced a working exploit (0%), far below Opus 4.8 and Mythos 5.
- Even so, it launched with cyber safeguards on by default, the same as Opus 4.7 and 4.8 and less strict than Fable 5's. It is part of the Cyber Verification Program.
What it means for businesses
For anyone using AI in customer service, process automation or coding, Sonnet 5 brings "Opus-level" capability at a "Sonnet" price. In practice:
- Retest your workflows: agents that used to stall halfway may now finish the job.
- Measure the real cost: with the new tokenizer, compare token usage before migrating.
- Keep a human in the loop for sensitive actions (customer emails, system changes), even with better prompt-injection resistance.
Sources
- Anthropic, "Introducing Claude Sonnet 5" (2026-06-30): https://www.anthropic.com/news/claude-sonnet-5
Want to test Claude or other AI models in your company's processes safely? KGA IT offers a free consultation: kga-it.com.