Anthropic shipped Claude Opus 4.7 today. If you're already using Claude for serious work, here's what changed, what didn't, and whether you need to do anything differently starting now.
What stayed the same — and that's the good news
Pricing didn't move. $5 per million input tokens, $25 per million output tokens. Access through the same API — just update the model string to claude-opus-4-7. If you're using the Claude app rather than the API, you're already on the new model automatically. No subscription changes, no new tiers. It's a clean upgrade.
Where Opus 4.7 earns its keep
The biggest jump is in complex coding tasks. Anthropic reports a 13% improvement on a 93-task coding benchmark compared to Opus 4.6 — and 4.6 was already strong. If you've been using Claude to build internal tools, automate workflows, or work through multi-step technical problems, you'll feel the difference in the harder tasks that used to need more hand-holding. Less back-and-forth, more getting it right the first time.
Vision also got a real upgrade. The model now handles images up to roughly 3.75 megapixels — more than triple what previous Claude versions could process. For businesses working with product photos, scanned documents, floor plans, or any image-heavy workflow, that extra resolution translates to more useful output from materials you're already using.
Finance teams, pay attention
Opus 4.7 posted state-of-the-art results on the Finance Agent evaluation — a benchmark testing economically valuable knowledge work, not trivia. The model also shows meaningful improvement in document reasoning, which is exactly what finance and legal teams actually need day-to-day. If you're using AI in your finance function for analysis, reporting, or document review, this version is worth a real test.
The one thing to watch
Better instruction following sounds like pure upside, but there's a catch: prompts tuned for Opus 4.6 may behave slightly differently here. The model adheres more precisely to what you literally ask, which means any prompt that relied on Claude "knowing what you meant" despite loose wording may need a refresh. Nothing catastrophic — but test your existing workflows before flipping the switch on anything production-critical.
What to do right now
Using the Claude app? Nothing — you're already on 4.7. Using the API with important workflows on Opus 4.6? Run your key prompts on 4.7 and check the output before switching the model string. Not yet on Claude's top tier? This upgrade makes the "cheaper models are good enough" argument harder to defend for complex work. Opus 4.7 hits harder where it counts.