Opus 4 and Sonnet 4 shipped together with extended thinking and parallel tool use, and Sonnet 4 reached the free tier.
The upgrade that matters for sellers is not raw capability, it is that the model can now think and use tools in the same pass — which is what a real account analysis requires, since it needs to look things up mid-reasoning rather than answer from a single snapshot.
Anthropic released two models. Opus 4 is the expensive, most capable one. Sonnet 4 is the everyday one, and it is what most people actually get.
Extended thinking with tool use is the headline change. Previously a model either reasoned or called a tool; now it can pause mid-reasoning, go and fetch something, and carry on. Parallel tool execution means it can fetch several things at once instead of queuing them.
The short version: the everyday Claude got materially better and can now look things up while it is still working through a problem.
Anthropic released Claude Opus 4 and Claude Sonnet 4 on 22 May 2025, adding extended thinking with tool use, parallel tool execution and improved memory for agents.
Price: Opus 4 at $15 per million input and $75 per million output tokens; Sonnet 4 at $3 and $15.
Availability: claude.ai, the Claude API, Amazon Bedrock and Google Cloud Vertex AI. Sonnet 4 available to free users; both models in Pro, Max, Team and Enterprise.
Reliability: Anthropic reported the models are 65% less likely than Sonnet 3.7 to take shortcuts on hard tasks.
Regions: Not stated.
Deprecation: Not stated — no retirement date given for the prior generation.
What it doesn't do: Opus 4 condenses its reasoning into thinking summaries roughly 5% of the time, and getting the raw chain of thought requires contacting sales for Developer Mode. The announcement's headline capability is extended thinking, not a larger context window — a distinction worth holding onto, because the two get conflated constantly.
SDR — Sonnet 4 reaching free users is the practical headline. If your team is not on paid seats, the quality gap between your drafting and a funded competitor's just narrowed considerably.
AE — Extended thinking with tool use is what makes multi-source deal analysis credible — the model can pull the account history, the competitor's latest filing and your notes into one pass rather than one at a time.
Manager — Opus 4's price is the constraint you will feel. At $15/$75 per million tokens it is roughly five times Sonnet, so decide deliberately which work justifies it rather than defaulting the team to the best model.
RevOps — Note the Developer Mode caveat: raw chains of thought require contacting sales. If your compliance posture requires full reasoning transcripts for auditability, the default summarised output will not satisfy it.
Extended thinking earns its cost on questions with several moving parts. Give it a whole quarter and make it defend more than one outcome.
Two minutes, once a week. What changed in AI, and what to run because of it.