Claude Sonnet 4.6 launched: 1M context, same pricing, stronger real-world automation
Original: Claude Sonnet 4.6 View original →
What Anthropic announced
On February 17, 2026, Anthropic introduced Claude Sonnet 4.6. The company positions it as a full upgrade across coding, computer use, long-context reasoning, agent planning, knowledge work, and design tasks. A key headline is a 1M token context window in beta.
Rollout details matter for teams already in production. According to Anthropic, Sonnet 4.6 became the default model for Free and Pro users in claude.ai and Claude Cowork. For API users, the important operational point is pricing: Sonnet 4.6 keeps Sonnet 4.5 pricing, starting at $3 input / $15 output per million tokens.
Why this release is notable
The practical claim is that more economically valuable tasks can now run on a Sonnet-tier model rather than requiring an Opus-tier model. Anthropic highlights improvements in consistency, instruction following, and coding performance from early-access feedback.
Another focus is computer use. Anthropic describes continued progress in letting models operate software through normal interfaces instead of bespoke API integrations. The announcement references OSWorld and OSWorld-Verified as indicators of the model family’s progress on realistic GUI tasks.
Safety and deployment implications
Anthropic also states that Sonnet 4.6 underwent safety evaluation via its system-card process, and reports outcomes comparable to or better than recent Claude models. It explicitly calls out prompt-injection resistance as a critical area for computer-use scenarios.
For engineering teams, this release is less about one benchmark and more about an architecture shift: if quality improves while price stays flat, workload routing can move from “premium-only” to “mixed-tier by task class.” That can reduce unit cost and latency for high-volume workflows. The tradeoff is governance complexity, especially around tool permissions and auditability when autonomous steps increase.
In short, Sonnet 4.6 appears to be a meaningful capability-per-dollar upgrade. The next step for most teams is straightforward: run side-by-side evals on internal task suites, then update routing policies only where reliability gains are clear.
Sources: Anthropic announcement, Claude Sonnet 4.6 system card, Hacker News discussion
Related Articles
Anthropic is not only shipping a stronger Claude model; it is splitting the same base capability into a broad Fable release and a restricted Mythos track. The package includes $10/$50 token pricing, 30-day safety retention, and automatic fallback to Opus 4.8 for some high-risk requests.
For months, Claude has been spontaneously telling users to go to sleep during active conversations, sometimes at 8:30 AM. Anthropic acknowledges the issue but hasn't identified the root cause, calling it 'a bit of a character tic.'
HN interest centered less on “Claude finds bugs” and more on the shape of a harness security teams can adapt for their own targets.