claude sonnet 5 is the default free model now. that changes the math

the headline that matters

On June 30, Anthropic launched Claude Sonnet 5. The benchmark charts are the part everyone screenshots. The part that matters for most people reading this is one line in the announcement: Sonnet 5 is now the default model on the Free and Pro plans.

Read that again. The model Anthropic describes as its most agentic Sonnet yet, the one that plans, uses browsers and terminals, and runs autonomously at a level that recently required bigger and more expensive models, is what you get when you open Claude without paying anything.

what "agentic" means in practice

Strip the jargon and it means this: instead of answering one message at a time, the model can hold a goal, break it into steps, use tools, and keep going until the job is done or it hits a wall. Earlier this year that behavior lived in premium tiers. Sonnet 5 brings it near Opus 4.8 territory, Anthropic's top model, at a fraction of the price.

Anthropic also reports it is a clear step up from Sonnet 4.6 on reasoning, tool use, coding, and knowledge work, and that safety testing showed a lower rate of undesirable behavior in agentic settings. Take vendor benchmarks with the usual salt. The direction is what counts.

the developer math

If you build on the API, the numbers are the story:

• Intro pricing: $2 per million input tokens, $10 per million output tokens, until August 31.
• After that: $3 in, $15 out.

For comparison, frontier-tier models bill multiples of that. If your product runs agents at any volume, this is the cost tier where experiments stop being scary. The intro window is real money too: if you have a heavy job to run, run it before September.

who should actually switch

• You pay for nothing and use the free plan: you already switched. You now have a near-frontier agent by default. Learn to give it multi-step tasks instead of single questions, that is where this model earns its keep.
• You pay for a mid-tier model API for agents: benchmark Sonnet 5 against it this month while the intro pricing holds. The cost-performance case is strong on paper.
• You run Opus-class workloads: do not switch blind. "Near Opus" is not Opus. Run your real tasks side by side and let the failures decide.

the honest caveat

Every model launch looks perfect for two weeks because everyone posts their best results. The real review happens on a Tuesday with your messy inputs. I am running mine now, and if Sonnet 5 falls short somewhere that matters, you will read it here.

The free plan just got dramatically better. That part is not hype. It is pricing strategy, and for once the strategy lands in your favor.