Anthropic released Claude Sonnet 5 on June 30, 2026, and it is the kind of release that actually matters for developers. Not because of benchmark theater, but because it hits a sweet spot: performance close to Opus 4.8 at roughly half the price.
If you use Claude Code, Codex, or any agentic coding tool, this is worth paying attention to.
What Is Actually New
Sonnet 5 is the most agentic Sonnet model Anthropic has shipped. It can plan, use tools like browsers and terminals, and run autonomously at a level that previously required Opus-class models.
The numbers tell the story. On BrowseComp (agentic search) and OSWorld-Verified (computer use), Sonnet 5 substantially outperforms Sonnet 4.6 and approaches Opus 4.8 at medium and high effort levels. It is not identical to Opus, but the gap is much smaller than it used to be.
Anthropic positions it as a cost-performance middle ground. Opus 4.8 costs $5 per million input tokens and $25 per million output tokens. Sonnet 5 launches at $2 per million input and $10 per million output through August 31, 2026. After that, pricing moves to $3 and $15 respectively. That is a meaningful difference for anyone running long agentic sessions where token costs add up fast.
What Changed Under the Hood
Sonnet 5 is substantially better than Sonnet 4.6 at reasoning, tool use, coding, and knowledge work. Anthropic's own safety assessments show lower rates of undesirable behaviors compared to its predecessor, and significantly reduced ability to perform cybersecurity tasks compared to Opus models.
The reduced cybersecurity capability is interesting. It suggests Anthropic is deliberately constraining Sonnet-class models on certain capabilities while keeping them strong on coding and agentic tasks. For most developers, this is a non-issue. You want a coding assistant, not a penetration testing tool.
Early access partners reported that Sonnet 5 handles complex multi-step tasks much more reliably than Sonnet 4.6. Tasks that previously required manual intervention or Opus-level models now complete successfully with Sonnet 5 at medium effort. That is a bigger deal than it sounds. Every time you have to interrupt an agentic session to correct a mistake, you lose flow state. Fewer interruptions means more productive coding time.
When to Use It vs. Opus
The decision framework is straightforward.
Use Sonnet 5 when you need fast, cost-effective agentic work: code reviews, refactoring, test writing, documentation, research tasks, and interactive coding sessions. The introductory pricing makes it cheap enough to run continuously without worrying about token burn.
Use Opus 4.8 when you need maximum capability on genuinely hard problems: complex architecture decisions, nuanced code generation across large codebases, tasks requiring high-level reasoning over long contexts, and anything where failure has significant consequences.
The effort level parameter is the key differentiator. Sonnet 5 at extra-high effort can match Opus 4.8 on some tasks. But it costs more tokens to get there, which reduces the cost advantage. The sweet spot is medium effort for most daily work, with extra-high reserved for genuinely difficult problems.
Migration Is Trivial
If you are already using Claude Code or Codex, switching to Sonnet 5 takes seconds. It is available as claude-sonnet-5 through the API. The Claude Code CLI and Claude Platform default to Sonnet 5 for Free and Pro plans starting now.
For API users, the model identifier is straightforward, and it is compatible with existing tool-use and agentic workflows. No code changes required. If you have existing scripts that reference claude-sonnet-4, updating the model identifier is a one-line change.
The Bigger Pattern
Sonnet 5 represents a broader trend in AI development: the capability floor keeps rising. What required a frontier model six months ago now runs on a mid-tier model at lower cost. This happens because training techniques improve, data quality increases, and competitive pressure from open-weights models forces everyone to ship better models at lower prices.
For developers, this is unambiguously good. Your coding tools get better and cheaper on a roughly quarterly cycle. The question is not whether to adopt these improvements. It is how to structure your workflow to take advantage of them without introducing instability.
Start with Sonnet 5 for daily work. Keep Opus available for the hard problems. And adjust as the models keep improving.
