Summary: Anthropic shipped Claude Sonnet 4.6 on February 17, 2026 and made it the default for free and Pro users, according to CNBC with a May 2025 knowledge cutoff. The article argues the default slot, not the model picker, determines most users’ experience and accelerates feedback at population scale. Practical use still needs grounding for post-cutoff events, packages, and APIs.
February 17, 2026: Anthropic ships Claude Sonnet 4.6, and the model steps straight into the default slot for free and Pro users. CNBC’s coverage adds one more detail — the build’s knowledge cutoff lands in May 2025.
That is twelve days on from Claude Opus 4.6 (February 5). Inside one fortnight, Anthropic has turned over both tiers — the flagship and the everyday workhorse.
The Real Fight Is Over the Default Slot
Most people never open the model picker. To them, Claude is whatever the default hands over. Dropping the newest Sonnet into that seat lifts the entire free and Pro population in a single motion — lived experience, word of mouth, and the usage data behind it all jump a generation together.
It is also the priciest seat to fill in a launch. Free-tier users are not handed a stripped-down older release; they get the same fresh build as Pro subscribers. That choice doubles as an acquisition play: the lowest-friction door in the house opens onto the newest capability.
No feedback loop in the industry spins faster than the default slot. Every conversation a free user starts becomes same-day telemetry on how the new model behaves, across every domain people bring to it. A flagship launch reveals a lab’s convictions; a default swap reveals what it wants every single user to feel immediately.
Two Lanes, One Matrix: Flagship and Workhorse
Set February’s two releases side by side and a clean product matrix emerges. Opus 4.6 claims the flagship lane — agent teams, a 1M-token context window, Claude in PowerPoint — built for heavy and enterprise use. Sonnet 4.6 plays the daily workhorse, absorbing most conversations and coding workloads.
For developers, that layering makes selection simple: demanding agent workflows point at Opus, routine tasks stay with Sonnet. The variable that actually drives the decision slides from “which one is stronger” to “how the cost structure gets arranged” — the look of a model market growing up.
Operationally, it is a load-bearing week: replacing the default for every free and Pro user in one stroke resets capacity, rate limits, and cost curves at the same moment. Pulling that trigger a dozen days after a flagship launch is a statement about confidence in the pipeline.
What a May 2025 Cutoff Means Day to Day
That cutoff falls roughly nine months short of the release date (February 2026). The world this model “knows” ends at mid-2025 — anything later, whether events, package versions, or API changes, simply never entered its training data.
What that means in practice:
- Time-sensitive work (new library versions, recent events) calls for retrieval or grounding tools layered on top of the model, not the model by itself
- When sizing up performance, separate “doesn’t know” from “does it poorly” before passing judgment on capability
- Teams moving over to Sonnet 4.6 should fold knowledge-dense scenarios into their regression testing
A cutoff like this makes the case for wiring fresh sources alongside the model as the rule rather than the exception — with retrieval built into the product itself instead of serving as an escape hatch. None of it is unique to Sonnet 4.6; every model leaves the factory with a cutoff. But a default model raises the stakes: questions that hinge on timing will now be answered at population scale by a model whose picture of the world stops in May 2025, unless fresher sources sit in front of it.