Claude Sonnet 5 is now the default, and defaults are the strategy

The most important number in any AI model release is not a benchmark. It is how many people get the model without asking for it. By that measure, Claude Sonnet 5 is Anthropic’s biggest launch of the year: it is now the default model for every Free and Pro user.

Per Anthropic’s announcement, Sonnet 5 is available everywhere, from the consumer apps through Claude Code and the developer platform, with introductory API pricing running through August 31. The company positions it as an agentic workhorse: multi-step tasks and tool use at close to flagship capability, at a fraction of flagship cost.

Defaults are strategy

Model launches usually compete for enthusiasts. Default swaps compete for everyone else, the millions who never open a model picker. Pushing an agent-grade model to the free tier raises the floor of what casual users experience, and it sharpens the wedge Anthropic is driving into the developer market, where the credit war among model providers is already redefining customer acquisition. Introductory pricing on an already-cheaper tier reads the same way: get workloads onto Sonnet 5 before the discount window closes. Defaults cut the other way too, when the thing switched on by default is other people: Meta made every public account opt-in to AI remixing before pulling the Muse Image @-mention feature three days later.

The economics that make a free-tier default possible

Handing an agent-grade model to every free user only works if serving it is cheap enough to give away, which makes the default swap a public statement about inference costs. Efficiency gains, from architecture, quantization and routing, have compressed the cost of near-flagship capability fast enough that the free tier can absorb it, and the introductory API pricing extends the same statement to developers: the floor price of capable inference keeps falling, exactly the dynamic pressuring application-layer margins from the other direction.

The default swap is also where distribution advantages get minted. A generation of casual users whose first agentic experience happens by default, without a model picker, forms habits around one vendor’s behavior and tone, and habits are the stickiest lock-in there is because nothing in a contract creates or dissolves them. That is the consumer mirror of the enterprise dynamics we mapped in the free-compute land grab: subsidize the entry point, own the habit, monetize the graduation. The model picker is becoming a power-user feature, and the vendors know that whoever owns the default owns the market segment that never opens it.

What to watch

Watch what happens to usage caps and Pro subscription value as the default improves, whether rivals respond by upgrading their own free tiers, and where the introductory pricing lands after August 31. Defaults, caps and price are now the real battleground; benchmarks are the sideshow.