Claude Sonnet 5: The Everyday Model That Might Actually Replace Opus for Most of Your Work

Anthropic just released Claude Sonnet 5, pitching it as their most capable mid-tier model yet — near-Opus-level results at Sonnet-level prices. If that holds up, it changes the everyday calculus for anyone building with Claude.

What Sonnet 5 Actually Is (And Isn't)

Let's get the positioning straight. Anthropic's model lineup has Opus at the expensive, heavy-duty end and Sonnet sitting in the middle — faster and cheaper, historically a little less capable. Sonnet 5 is meant to close that gap dramatically.

According to Claire Vo's review on Lenny's Newsletter — where she ran 64 generations across real product tasks as a structured benchmark — Anthropic is claiming Sonnet 5 performs close to Opus 4.8 on the kinds of work most people actually do: writing PRDs (product requirements docs), debugging, and one-shot design generation. Not quite at Opus's ceiling, but close enough that she suspects "most of us are not going to notice the difference."

Pricing at launch: $2 per million input tokens and $10 per million output tokens, at least through the end of the summer, after which it's expected to rise. If you want to stress-test it at the lowest rate, now is the window.

The Thesis: Stop Defaulting to Opus

Here's the thing that should actually change your habits: an Anthropic team member told Claire that people are "abusing Opus" — meaning designers and builders are reflexively reaching for the most powerful (and most expensive) model for tasks that a mid-tier model handles just fine.

Sonnet 5 is designed to be your everyday driver, not your special-occasion tool. For a designer who is vibe-coding — prompting your way through an app build without writing raw code — this matters practically. If you're running lots of iterations (and you will be), paying Opus prices on every single generation adds up fast. Sonnet 5 gives you a credible alternative that doesn't feel like a significant downgrade.

Where Sonnet 5 Genuinely Pulls Ahead

Two areas stand out from the benchmarking:

Agentic tool use. "Agentic" just means the model can run longer sequences of actions on its own — chaining steps together rather than waiting for you to prompt each one. Sonnet 5 with its "extra high reasoning" mode enabled gets noticeably closer to Opus on these longer-running tasks than its predecessor, Sonnet 4.6, did. If you're using Claude inside tools like Cursor or Claude Code to build multi-step features, this is the capability that matters.

Computer and browser use. This is Claude's ability to actually operate software — clicking, navigating, filling forms — rather than just generating text. Sonnet 4.6 sat around an 80% pass rate on these tasks. Sonnet 5 pushes past that, approaching Opus territory at a lower cost. For designers experimenting with AI agents that can do research, scrape references, or interact with other tools, this is a meaningful jump.

A Smarter Way to Evaluate Any New Model

The most practically useful thing in Claire's review isn't the Sonnet 5 verdict itself — it's the benchmarking approach she built to reach it. Instead of running a one-off "vibe check" (prompting the model once and forming an impression), she used Claude Code to build a repeatable eval set: a defined list of tasks she cares about — PRD writing, bug-solving, one-shot design — tested consistently across models.

One detail worth stealing: Claude Code stores all your past sessions on your desktop. You can ask Claude to review that history and use it as context when designing new tests or workflows. She notes this works with other tools like Codex too. For a designer, this means your AI tool can learn from the pattern of work you've already done together — surfacing smarter suggestions, not just responding to your latest prompt in isolation.

If you're building with AI regularly, setting up even a lightweight personal benchmark — three or four prompts that represent your actual workflow, run blind across models — will tell you far more than any headline benchmark.

What to Keep in Mind

Claire ran her evals live, with some scores still coming in during the episode itself. That's honest, but it also means the full picture is still developing. Sonnet 5 is new, and real-world performance across the messy variety of design and product tasks will take weeks of community use to fully understand.

Also worth noting: the launch pricing is temporary. If the cost advantage is part of your reasoning for switching to Sonnet 5, build that assumption into your planning knowing it's likely to shift.

The grounded takeaway: Sonnet 5 looks like a genuine upgrade for the kind of iterative, agentic work that designers doing their own building actually do. It's probably not worth obsessing over whether it matches Opus on every dimension. It's worth trying it as your default — and paying attention to where it falls short for your specific tasks, rather than trusting any single review's conclusions as your own.