Anthropic shipped Claude Opus 4.7 on April 16, 2026. It is available now through Anthropic’s API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry, at the same base price as its predecessor: $5 per million input tokens and $25 per million output tokens.

The headline number is the coding benchmark. On SWE-bench Verified, the standard test for autonomous software engineering, Opus 4.7 scores 87.6 percent. That is up from 80.8 percent for Opus 4.6 and puts it ahead of GPT-5.4 and Gemini 3.1 Pro on the same test. On Rakuten’s internal production task benchmark, Opus 4.7 resolves three times more real-world tasks than Opus 4.6.

For businesses using Claude in software development workflows, the jump from 80.8 to 87.6 on SWE-bench is not a marginal improvement. It represents a meaningful step toward agents that can handle production code, not just greenfield problems.

What Changed in 4.7

Three areas saw the biggest upgrades.

Coding and agentic work. The 13 percent lift on coding benchmarks compounds with a new ability to verify its own output. Claude Opus 4.7 can double-check its reasoning before finalising a response, which matters most in long-running agentic tasks where early errors compound. New task budget controls give developers explicit handles over how much effort the model spends on a given step, which translates to more predictable costs in production.

Vision. Maximum image resolution jumped from 1,568 pixels on the long edge (roughly 1.15 megapixels) to 2,576 pixels (roughly 3.75 megapixels). That is more than three times the visual capacity of prior Claude models. Teams processing documents, charts, diagrams, or screenshots will see noticeably better extraction accuracy as a result.

Reasoning control. Anthropic introduced a new effort level called “xhigh” that sits between the existing “high” and “max” settings. This gives developers finer-grained control over the tradeoff between reasoning depth and response speed on hard problems, without having to jump straight to the most expensive compute tier.

One Catch on Pricing

The API rate is unchanged, but Anthropic also shipped a new tokenizer with this model. The same input text can now map to roughly 1.0 to 1.35 times as many tokens depending on content type. For teams running high volumes, total cost may rise even with stable per-token rates. Worth modelling out before switching production workloads.

Where Opus 4.7 Sits in Anthropic’s Lineup

Anthropic is transparent about the positioning. Claude Mythos Preview, announced earlier this month, remains Anthropic’s most capable model by a wide margin. On USAMO 2026, Mythos scored 97.6 percent versus Opus 4.7’s much lower result. But Mythos is restricted to a small group of tech and cybersecurity companies through Project Glasswing and is not publicly accessible.

Opus 4.7 is the most capable model available to general enterprise customers right now. For the vast majority of teams, the question is not Opus vs Mythos. It is whether Opus 4.7 is meaningfully better than what they are currently running, and in software engineering and document understanding tasks, the answer appears to be yes.

What This Means for Business

If your team is using Claude for any of the following, a direct comparison test is worth running this week:

Code review and generation. The three-times improvement on production task resolution suggests Opus 4.7 will close more real tickets, not just score better on academic benchmarks. Teams using agentic coding pipelines should see fewer rollbacks from incorrect patches.

Document processing. The resolution upgrade is directly relevant to anyone extracting data from invoices, reports, contracts, or dashboards. Higher-fidelity image understanding means fewer OCR errors and less need for pre-processing pipelines.

Long-running agents. The self-verification capability and task budgets make Opus 4.7 more suitable for multi-step autonomous workflows where errors need to be caught before they propagate. If your current Claude implementation breaks on long tasks, this release is a practical reason to re-evaluate.

Cost planning. The tokenizer change is worth a close look for anyone running large volumes. Benchmark your current prompts against the new tokenizer before committing to a migration.

The broader pattern here is Anthropic converging on a clear two-tier model strategy: a restricted frontier for cutting-edge research and security, and a commercially available flagship for business use. Opus 4.7 is that commercial flagship. For data and AI teams planning their tool stack for the rest of 2026, it now sets a new baseline.

---

Enterprise DNA helps businesses build the data skills and AI systems they need to stay competitive. Whether you are upskilling your team through EDNA Learn or deploying AI agents with Omni, we can help you move from evaluation to production faster.