Anthropic released Claude Sonnet 5 on June 30, 2026. The headline isn’t just that it’s better than the last Sonnet. It’s how close it now sits to Anthropic’s own top-tier model, at a fraction of the price.
Sonnet has always been the workhorse tier. It’s the model most developers actually build on day to day. Anthropic says this version narrows the gap with Opus 4.8 substantially. It stays meaningfully cheaper than Opus too.
What’s New
- Anthropic says Sonnet 5 is a significant upgrade over Sonnet 4.6. Agentic tasks, reasoning, tool use, coding, and everyday knowledge work all improved, according to Anthropic’s own benchmark comparisons
- According to Anthropic, using higher effort levels allows Sonnet 5 to match Opus 4.8 on some benchmarked tasks. That gives developers a cost-to-performance dial instead of a hard choice between two separate models
- It launches today across every plan. Free and Pro users get it as the default model. It’s available now on Max, Team, and Enterprise too, plus inside Claude Code and through the Claude API
- Introductory pricing runs $2 per million input tokens and $10 per million output tokens through August 31, 2026. After that, it moves to standard pricing at $3 input and $15 output
- Anthropic says early access partners reported that Sonnet 5 was more capable of completing complex, multi-step tasks that previous Sonnet models often left unfinished. The company also says partners observed the model checking its own work more frequently without being explicitly prompted
- Sonnet 5 uses an updated tokenizer, the same kind of change made with Opus 4.7. That means the same input can turn into somewhat more tokens than before. Anthropic set the introductory pricing to keep the switch roughly cost-neutral despite that
Why It Matters
Anthropic has spent the last several model generations pulling ahead on its Opus tier. Sonnet stayed the budget option. The release suggests Anthropic is narrowing the gap between its mid-tier and flagship models. Instead of a big capability gap between “the good model” and “the cheap model,” Anthropic is now selling a single model that can flex between both roles. How much it flexes depends on how much effort you tell it to spend.
That matters for anyone building on Claude, not just people chasing benchmark scores. A model that scales its own performance up or down based on task difficulty is a more practical tool than two separate models with a hard line between them. It also puts pressure on how much anyone needs to reach for Opus by default. Sonnet 5 now covers more of that ground on its own.
Anthropic also built in cyber safeguards by default this time. Sonnet 5 tested somewhat stronger at technical tasks than its predecessor. It showed a slightly higher partial success rate on a Firefox exploit-development test, though it still couldn’t produce a full working exploit in that evaluation. Because of that increase, real-time safeguards that detect and block dangerous cyber usage now ship active out of the box. These are the same protections already used on Opus 4.7 and 4.8.
Should You Care?
If you use Claude at all, yes, directly. Sonnet 5 is now the default model on Free and Pro plans. Most people will simply start using it without changing anything.
If you build on the Claude API, the pricing window is worth knowing about. The introductory rate holds through August 31, 2026. After that, it moves to standard pricing. If you’re planning a larger migration or usage ramp-up, timing it inside that window is the practical thing to note here.
Source: Anthropic: Introducing Claude Sonnet 5
Recent Anthropic Releases: Claude Tag