What is Claude Opus 4.7 and what changes from 4.6
On April 16, 2026, Anthropic released Claude Opus 4.7, the direct successor to Opus 4.6 and the new flagship model in the Claude family. It is a hybrid reasoning model: it combines immediate response capabilities with deep reasoning on demand, optimizing the balance between output quality and latency.
The context window remains at 1 million tokens — one of the highest values among large models available on the market. This allows analyzing entire code repositories, multiple contracts, complex financial documents or meeting transcripts without losing the thread of reasoning.
The most important structural technical change is the updated tokenizer: the same text input generates between 1.0x and 1.35x more tokens compared to Opus 4.6. This has direct implications for API costs and budget planning. We cover this in detail in the Claude Opus 4.7 vs Opus 4.6 article.
Another notable technical innovation is the new effort control level: `xhigh`, placed between `high` and `max`. It allows developers to calibrate the balance between reasoning depth and response latency with greater precision — useful for enterprise workflows where response time is a critical constraint.
Official benchmarks: the numbers that matter
Anthropic published a series of benchmarks with industry partners that show substantial improvements over Opus 4.6.
On coding, CursorBench measures the completion of real programming tasks: Opus 4.7 reaches 70% against Opus 4.6's 58%. Rakuten-SWE-Bench, which measures solving software development tasks in production environments, shows that Opus 4.7 solves 3x more tasks than its predecessor. CodeRabbit, specialized in automated code review, records a more than 10% improvement in code problem recall.
In vision and document analysis, the most relevant result comes from XBOW, measuring visual acuity — the ability to correctly interpret images and visual documents: Opus 4.7 reaches 98.5% against 54.5% for Opus 4.6. A 44-percentage-point jump that opens new scenarios for processing scanned invoices, paper contracts, technical diagrams and chemical structures. Image support reaches up to 2,576 pixels on the long side (about 3.75 megapixels), more than three times the limit of previous Claude models.
For the legal sector, Harvey reports 90.9% accuracy on BigLaw Bench, the reference benchmark for complex legal tasks. In financial analysis, the General Finance module scores 0.813 vs 0.767 for Opus 4.6, with Hex reporting superior performance on missing data handling.
For multi-step workflows, Notion Agent records +14% compared to Opus 4.6, while Databricks OfficeQA Pro shows 21% fewer errors on enterprise document analysis tasks.
Want to adopt Claude Opus 4.7 in your enterprise?
30 minutes to discuss your specific case.
Pricing and availability: where to find it and what it costs
Claude Opus 4.7 is available from release day on all major channels: Claude.ai in Pro, Max, Team and Enterprise plans; direct Claude API; Amazon Bedrock; Google Cloud Vertex AI; Microsoft Foundry. The API model ID is `claude-opus-4-7`.
Pricing remains unchanged from Opus 4.6: $5 per million input tokens, $25 per million output tokens. At first glance this seems like good news — but the updated tokenizer (which generates up to 35% more tokens from the same text) means the same document will cost slightly more to process than before. For heavy API users, the impact needs to be calculated based on your actual volume.
For enterprises using Claude through managed plans (Team or Enterprise), pricing doesn't change directly. What changes is the quality of responses obtained, which justifies transitioning to the new model. For those evaluating available plans, the guide on what Claude costs for enterprises is the updated reference.
Availability on Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry means enterprises with existing cloud infrastructure can integrate Opus 4.7 without migrating to the direct Anthropic API — important flexibility for IT teams with pre-defined architecture constraints.
Primary use cases according to Anthropic
Anthropic has identified five primary application areas for Opus 4.7, all supported by published benchmarks.
Advanced coding and code review: the combined CursorBench 70%, Rakuten 3x and CodeRabbit +10% defines Opus 4.7 as the reference model for development teams working on complex codebases. Not just code writing, but pull request analysis, bug identification and structural refactoring.
AI agents for multi-tool workflows: the +14% from Notion on multi-step workflows and the new `xhigh` effort control level make Opus 4.7 particularly suited for agents orchestrating multiple tools in sequence. For those building agentic solutions, the Claude Opus 4.7 for AI agents article covers the technical details.
Multi-day enterprise workflows: complex spreadsheets, slide decks, policy documents — tasks requiring coherence over long time horizons and large volumes of context.
Computer vision: chemical structures, technical diagrams, engineering blueprints. The jump in visual acuity (98.5% vs 54.5%) makes Opus 4.7 a practical tool for automated analysis of non-natively digital documents.
Document analysis and reasoning: contracts, financial statements, due diligence reports. The combination of wide context window and improved reasoning makes it suitable for tasks where document understanding must be precise and verifiable.
What it means for enterprises using Claude today
If your company already uses Claude — via API, Claude.ai or a Team/Enterprise plan — Opus 4.7 is a natural upgrade for tasks where quality matters more than cost per token.
The practical advice is to start evaluating Opus 4.7 on your most critical workflows: those where an error is costly, where document complexity is high, or where output feeds important decisions. The benchmarks indicate real and measurable gains — but validation on your specific use case remains necessary.
For API users, the updated tokenizer requires re-examining cost estimates. It's not a marginal change: up to 35% more tokens from the same text impacts budgets at high volumes. The Claude Opus 4.7 vs Opus 4.6 article analyzes this in detail with practical examples.
For development teams, the Rakuten benchmark (3x tasks solved in production) is the most significant data point. It's not a synthetic benchmark: it measures the ability to solve real problems on real codebases. For those using Claude in the software development cycle, testing Opus 4.7 directly is worthwhile. The Claude Opus 4.7 for coding article covers the details.
Maverick AI supports enterprises in adopting and evaluating Claude Opus 4.7, from choosing the right plan to integrating with existing systems. If you want to understand how Opus 4.7 fits into your AI architecture, talk to our team.