On April 16, 2026, Anthropic released Claude Opus 4.7 (with a follow-on feature update on May 4 that added the new "xhigh" effort tier). It is the newest version of its flagship model. The headline upgrades are real, and a couple of them matter more for small business than the benchmark numbers suggest. Here is what is actually different and where you should care.
What Got Better
Opus 4.7 is a step up across the board on the agentic, long-horizon tasks that increasingly define how a business uses AI. The most-cited number is coding, but the same gains show up on multi-step research, data work, and tool use.
Source: Anthropic. SWE-bench Pro climbed from 53.4% to 64.3% in the same release.
Three other changes do not look like much in a benchmark chart, but they change the economics of running an agent for a real business.
- Task budgets. You can now give Opus 4.7 a rough token budget for a full agentic loop, covering thinking, tool calls, results, and the final answer. In plain English: you can finally cap what a single agent run costs you.
- Adaptive thinking. The model decides how much "thinking" a task needs. Easy questions return fast. Hard ones get more compute. You stop paying for deep reasoning on shallow questions.
- New "xhigh" effort tier. A new effort level sits between "high" and "max" so you can dial up reasoning on the few tasks that need it without paying max-tier prices on everything else.
What Else Shipped
Two more upgrades round out the release.
- High-resolution vision. Opus 4.7 is the first Claude model to accept images up to 2576px (3.75MP), up from 1568px. That is the difference between "I can read this screenshot" and "I can read the fine print on this contract photo."
- 1M-token context, 128K output. Held over from 4.6, but now the default. You can hand it an entire client folder and ask it to summarize the last six months of work.
The Honest "Should I Care?" Read
If you talk to Claude.ai every day
You are already on it. Claude.ai upgraded automatically. Expect slightly sharper answers, faster responses on simple questions, and the new vision quality on screenshots.
If you build agents or workflows on the API
This is the release that matters. Task budgets and adaptive thinking together solve the single biggest objection small businesses raise about agents: "what if it runs all night and costs me $400?" You can now answer that with a hard ceiling per task.
If you have not started yet
The model layer is no longer the bottleneck. The hard part is picking one workflow and wiring it. We have been saying this for a year. Opus 4.7 just shrinks the excuse window further.
The Bottom Line
Opus 4.7 is the first Claude release where the most important features are not about raw intelligence. They are about predictability and cost control. For a small business, those are the two things that have always blocked production use. The frontier got smarter again, and it also got safer to run.
Sources
- Anthropic, "Introducing Claude Opus 4.7," official launch announcement, May 4, 2026. anthropic.com
- Anthropic, "What's new in Claude Opus 4.7" (developer docs covering task budgets, adaptive thinking, effort tiers, and vision). platform.claude.com
- Amazon Web Services, "Claude Opus 4.7 is now available in Amazon Bedrock," April 2026 launch coverage. aws.amazon.com
- GitHub Changelog, "Claude Opus 4.7 is generally available," April 16, 2026. github.blog
Want a model upgrade like 4.7 to actually translate into a result for your business?
We help small businesses pick the right model, wire the right workflow, and put a cost ceiling on every run.