Claude Opus 4.7 review: Is the upgrade to 4.8 worth it?
Release date: 16 April 2026 | Status: Active | Licence: Closed
Anthropic's flagship model had one of its shortest reigns yet. Claude Opus 4.7 landed on 16 April 2026, and by late May it was already replaced by Opus 4.8, roughly six weeks at the top.
For a business team, that raises a practical question rather than a technical one. If you've already wired 4.7 into a product or workflow, do you have to do anything? And if you're starting fresh, which one do you point your code at?
The short answer: the two models cost the same and share the same context window, so the decision really comes down to coding performance and how soon you think the older model gets retired. Here's where it sits.
Benchmarks at a glance
| Metric | Opus 4.7 | Opus 4.8 | Delta |
|---|---|---|---|
| SWE-bench Pro | 63.8% | 69.2% | +5.4 pts |
| MMLU | 89.2% | 89.8% | +0.6 pts |
| Context window | 1M (beta) | 1M (beta) | , |
| Price (input) | $5.00 / 1M | $5.00 / 1M | , |
| Price (output) | $25.00 / 1M | $25.00 / 1M | , |
A note on those numbers before you lean on them. Opus 4.8's 69.2% on SWE-bench Pro is the figure reported by independent benchmark trackers. The 4.7 figure in the table (63.8%) is lower than what we've seen elsewhere, most sources put 4.7 closer to 64.3%, which would make the real coding gain about 4.9 points rather than 5.4. The MMLU row should be treated with even more caution: we couldn't find MMLU scores published for either model, and most outlets have stopped reporting MMLU for frontier models, so treat 89.2% and 89.8% as unconfirmed.
One more correction worth flagging: the table lists the 1M context window as "(beta)", but Anthropic moved the 1M window to general availability on 13 March 2026, before either of these models shipped. So it's GA, not beta, on both.
The case for upgrading
The coding gain is the part that matters. A few points on SWE-bench Pro might sound trivial, but in coding work that range is usually where a model starts handling the harder cases, messy specs, edge conditions, changes that span several files at once. Reporting on 4.8 frames coding and agentic work as its headline improvement, and Anthropic points to better honesty too, with the model far less likely to wave through a flaw in code. If you're using Opus for software engineering, 4.8 is the one to be on.
The general-knowledge difference is another story. Even taking the unconfirmed MMLU figures at face value, a gap that small is noise for most uses. You won't feel it in everyday Q&A or document analysis.
The case for staying
There isn't much of one. The only real reason to hold on 4.7 is if something in your integration broke when you tried 4.8, say, parsing code that's sensitive to small shifts in how responses are structured. We didn't hit any breaking changes in our own testing, and Anthropic positioned 4.8 as a drop-in upgrade, though we couldn't find an explicit confirmation that the API schema is byte-for-byte identical. If you've got brittle parsing, test before you flip the switch.
One caveat: identical pricing
The base price is the same on both models, $5.00 input and $25.00 output per million tokens, so there's no money to be saved by staying on 4.7. Worth knowing: 4.8 also has a faster, pricier tier ($10/$50 per million tokens) that the original table doesn't mention, so "identical pricing" holds for the standard tier only.
Anthropic hasn't cut the price of the older model either. That's our read, not their statement, but it usually points to a model heading for deprecation. If you're building something new, target 4.8 explicitly.
Verdict
Opus 4.7 was a strong model for the few weeks it led. As of June 2026 it's been overtaken, and the upgrade path is clear: move to 4.8 unless you've got a specific technical blocker holding you back.
Score: 8.0 / 10 (at time of release) / 7.0 / 10 (relative to current options)