Claude 4.6 rollout and outages reported

Anthropic spent the past week pushing Claude 4.6 deeper into products and workplaces, then spent April 15 and April 16 fixing outages and answering complaints that the model now feels weaker. (platform.claude.com) (status.claude.com) (axios.com) Anthropic’s Claude 4.6 docs say Opus 4.6 and Sonnet 4.6 both support a 1 million token context window, adaptive thinking, and the company’s existing application programming interface features. Opus 4.6 offers 128,000 maximum output tokens, while Sonnet 4.6 offers 64,000. (platform.claude.com) Anthropic said adaptive thinking lets Claude decide when and how much to “think,” and its docs now recommend using an effort setting to control depth. The same page says a new “max” effort level is available on Opus 4.6, while Sonnet 4.6 added the effort parameter for the first time. (platform.claude.com) That setting sits near the center of the backlash. Axios reported on April 16 that users across online forums said Claude felt worse, and Claude Code creator Boris Cherny said the product now defaults to “medium effort” to balance intelligence with speed. (axios.com) Anthropic has also been widening where Claude 4.6 shows up. In its February 17 launch post, the company said Sonnet 4.6 became the default model in Claude.ai and Claude Cowork for Free and Pro users, while keeping Sonnet 4.5 pricing at $3 input and $15 output per million tokens. (anthropic.com) On April 9, Anthropic moved Claude Cowork into general availability on paid plans and added enterprise controls around it. The company said those controls include role-based access, group spend limits, usage analytics, and expanded OpenTelemetry monitoring. (claude.com) (support.claude.com) Then the service faltered. Anthropic’s status page says it began investigating increased errors on Claude.ai, the application programming interface, and Claude Code at 14:53 Coordinated Universal Time on April 15, and said at 15:40 Coordinated Universal Time that the application programming interface had recovered while Claude.ai remained down. (status.claude.com) The same status page said login success rates for Claude.ai, including through Claude Code, had stabilized by 16:29 Coordinated Universal Time, and the April 15 incident was later marked resolved at 17:42 Coordinated Universal Time. CNBC reported roughly 6,000 Downdetector reports at 10:42 a.m. Eastern Time, falling to about 2,000 by 1:12 p.m. Eastern Time. (status.claude.com) (cnbc.com) A second incident followed early April 16. Anthropic’s status page says Opus 4.6 had an elevated error-rate outage from 23:03 Pacific Time on April 15 to 00:26 Pacific Time on April 16, or 06:03 to 07:26 Coordinated Universal Time. (status.claude.com) Anthropic’s own pitch for 4.6 is that users can trade speed, cost, and performance more directly. The complaints now are that some of those tradeoffs may already be happening by default. (platform.claude.com) (axios.com)