Anthropic has launched Claude Sonnet 5, its most agentic mid-tier model yet. It runs close to the flagship Opus 4.8 on many tasks, but costs less than half as much.
Anthropic said on June 30, 2026 that Sonnet 5 is available today across every plan. The company built it to act, not just answer. It can make plans, drive browsers and terminals, and run on its own for long stretches. That kind of work needed bigger, pricier models only a few months ago.
The pitch is simple. Sonnet 5 offers near-flagship performance at a mid-tier price. It lands close to Opus 4.8, Anthropic’s most capable model, on reasoning, tool use, coding, and knowledge work. It clearly beats its predecessor, Sonnet 4.6. And it costs far less than Opus to run.
Cheaper agents, on purpose
Price sits at the centre of this launch. Sonnet 5 starts at $2 per million input tokens and $10 per million output tokens. That introductory rate holds until August 31, 2026. After that it moves to $3 and $15. Opus 4.8, by contrast, costs $5 and $25. TechCrunch framed the model as a cheaper way to run agents, and that is the point.
The timing matters. Companies rushed to deploy AI agents, then recoiled at the bills. Agents loop, call tools, and burn tokens fast. A model that gets close to Opus quality for a fraction of the cost speaks directly to that pain. It also speaks to a market hunting for savings after enterprise AI bills ballooned.
There is a catch in the small print. Sonnet 5 uses a new tokenizer, so the same text can map to up to 1.35 times more tokens than before. Anthropic set the introductory price so the switch stays roughly cost-neutral. The headline rate looks low, but the token count can climb.
How good is it?
On Anthropic’s own benchmarks, Sonnet 5 marks a clear step up from 4.6 without quite catching Opus. On an agentic coding test it scored 63.2 per cent, against 69.2 per cent for Opus 4.8 and 58.1 per cent for Sonnet 4.6, according to early reporting. On one knowledge-work benchmark it edged ahead of Opus. Anthropic also offers an “effort” dial, letting developers trade cost for accuracy between the two models.
Early testers told Anthropic the model finishes complex jobs where older Sonnets gave up, and that it checks its own output without being asked. Those claims come from the company’s launch material, so they deserve the usual caution. Independent testing will tell the real story.
Safer, with a cyber caveat
Anthropic says Sonnet 5 behaves better than 4.6 on safety. It refuses malicious requests more often and resists prompt-injection attacks, where hidden instructions try to hijack an agent. It also hallucinates and flatters less. On an automated audit of misaligned behaviour, it scored safer than 4.6, though worse than Opus 4.8 and the Mythos preview.
Cybersecurity is the sharper point. Anthropic did not train Sonnet 5 for cyber tasks, and it performs poorly at building software exploits. In a test run with Mozilla on the Firefox browser, the model never produced a working exploit. Even so, Anthropic shipped it with real-time cyber safeguards on by default, the same ones used on Opus 4.7 and 4.8. Those guardrails stay lighter than the ones around Fable 5, its locked-down public model.
A discount with a strategy behind it
The low price is not charity. Anthropic is racing rivals for developers, and a capable, affordable agent model is how you win them. The company also writes much of its own code with Claude, so a better, cheaper Sonnet helps its own engineers too. It is also moving toward a planned public listing, where revenue growth and developer reach both count.
The wider context is cost. Running agents around the clock can rack up eye-watering bills, and Anthropic has set out ambitious revenue targets to fund its model work. Sonnet 5 is its answer to both. Push capability down the price curve, keep developers inside the ecosystem, and let the effort dial handle the rest.
Claude Sonnet 5 is live now in Claude’s apps, Claude Code, and the API, with higher rate limits across the board. For most developers, the question is no longer whether the model is clever enough. It is whether it is cheap enough to run all day. Anthropic is betting the answer is finally yes.
Get the TNW newsletter
Get the most important tech news in your inbox each week.