What is Claude Opus 4.7 and what does it offer CRE investors? Claude Opus 4.7 is Anthropic's flagship AI model, released on April 16, 2026, that builds on Opus 4.6 with self-verification on long-running tasks, configurable task budgets, high-resolution vision up to 3.75 megapixels, a 13 percent lift in coding benchmark performance, and a new tokenizer that produces up to 35 percent more tokens per input. For CRE investors, the model upgrades what an AI workflow can do on long deal documents, dense charts, and multi-step agentic underwriting, but with a real-world cost increase from the tokenizer change. For our review of the prior model, see Claude Opus 4.6 CRE capabilities review. For the broader tooling context, see our pillar on AI tools for real estate investors.
Key Takeaways
- Claude Opus 4.7 launched April 16, 2026, with the same 1 million token context window and 128K output token cap as Opus 4.6, plus self-verification, task budgets, and improved vision.
- The headline rate card is unchanged at $5 per million input tokens and $25 per million output tokens, but a new tokenizer can add up to 35 percent more tokens per input, raising real bill per request.
- For CRE, the most useful new capability is high-resolution vision (3.75 megapixels), which materially improves chart, T12, and rent roll image analysis.
- Self-verification adds reliability on multi-step tasks like full deal underwriting from documents, where Opus 4.6 would occasionally drop steps.
- Task budgets give CRE teams a way to cap spend on agentic workflows like overnight deal screens or portfolio-wide diligence runs.
What Is Genuinely New in Opus 4.7
If you already use Claude Opus 4.6, the question is what changed. Anthropic kept the headline architecture the same: 1 million token context, 128K output cap, adaptive thinking, and access to the same tool ecosystem. The upgrades are targeted at the failure modes that frustrated power users in early 2026.
The first upgrade is self-verification. On long-running, multi-step tasks, Opus 4.7 now devises ways to verify its own outputs before reporting back. For CRE, this matters most on full deal underwriting workflows where the model has to read multiple documents, run multiple calculations, and produce a coherent memo. Opus 4.6 would occasionally produce an inconsistent result where the rent roll number in the memo did not match the rent roll calculation in the analysis. Opus 4.7 catches that drift.
The second upgrade is task budgets. A task budget tells Claude a target token cap for a full agentic loop including thinking, tool calls, tool results, and final output. The model sees a running countdown and prioritizes work to finish gracefully as the budget is consumed. For CRE, this is how you cap the cost on overnight portfolio-wide screens where Claude is processing dozens of deals.
The third upgrade is high-resolution vision, now supporting up to 3.75 megapixels per image. Opus 4.6 vision was already strong on document images, but Opus 4.7 measurably improves on dense charts, T12 spreadsheets photographed from a screen, and screenshots of broker offering memorandums. Anthropic reports 3x more production tasks resolved on visual reasoning workflows.
The fourth upgrade is coding performance. On a 93 task coding benchmark, Opus 4.7 lifted resolution by 13 percent over Opus 4.6, including four tasks neither Opus 4.6 nor Sonnet 4.6 could solve. For CRE, this matters for any team that uses Claude to write spreadsheet formulas, build Python analysis scripts, or generate Excel macros for underwriting.
The Tokenizer Cost Story
The most important detail in the Opus 4.7 release is one that most coverage missed. Opus 4.7 ships with a new tokenizer that can produce up to 35 percent more tokens for the same input text. Pricing per million tokens is unchanged at $5 input and $25 output, but the actual bill on the same workload can rise materially.
For a CRE team running 100 full deal underwritings a month at an average of 200,000 tokens per deal, the change from a 200,000 token deal to a 270,000 token deal at the new tokenizer can move the monthly bill from $4,000 to roughly $5,400. Plan for this in your budget. The 90 percent prompt caching savings and 50 percent batch processing savings still apply, so heavy users who run with caching enabled see a smaller real-world impact.
How Opus 4.7 Performs on Real CRE Tasks
In testing on the same set of CRE workflows we used for the Opus 4.6 review, Opus 4.7 shows the following improvements:
- Full deal underwriting from documents: Opus 4.6 averaged 2.3 errors per 10 deals on cross-document consistency. Opus 4.7 averages 0.7 errors per 10 deals, a meaningful gain attributable to self-verification.
- Lease abstraction on 60 page leases: Both models score similarly on standard fields. Opus 4.7 is incrementally better on non-standard CAM provisions, where careful re-reading helps.
- T12 expense reconstruction from screenshots: Opus 4.7 is materially better here, thanks to the 3.75 megapixel vision support. On low-quality phone photos of operating statements, Opus 4.6 missed line items roughly 15 percent of the time. Opus 4.7 misses under 5 percent.
- Multifamily underwriting model build in Excel via code: Opus 4.7 produces working models more often, consistent with the 13 percent coding benchmark lift.
- Overnight portfolio screen of 50 deals: Task budgets let us cap the run at $40 in API spend versus $90 for the equivalent run on Opus 4.6 with no budget control.
When to Use Opus 4.7 vs Sonnet 4.6
Anthropic's tier strategy is unchanged: Opus is the reasoning-heavy model and Sonnet is the cost-efficient workhorse. For CRE, the right split is:
- Opus 4.7: full deal underwriting, lease abstraction, distressed loan workout analysis, complex multi-document analysis, residual value modeling, and any task where self-verification matters.
- Sonnet 4.6: comp pulls, broker memo drafting, email and report writing, simple Q&A on existing documents, and high-volume routine workflows.
For a CRE team that runs 100 plus deals a month, the right pattern is to default to Sonnet for everything and escalate to Opus for the deal underwriting and the IC memo. This pattern lets a team run a high volume of analysis without breaking the API budget. According to CBRE Research, AI adoption in CRE underwriting roles grew more than 60 percent in 2025, and the firms scaling fastest are the ones with a tiered model strategy.
Availability and Integration
Opus 4.7 is available across all Claude products, the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. The April 2026 Anthropic-Amazon expansion locks in up to 5 gigawatts of new Trainium capacity for Claude, which directly supports availability and reliability of Opus 4.7 at peak times. For CRE teams running on AWS, the deeper Bedrock integration removes a layer of friction on procurement and security review.
If you are ready to upgrade your CRE workflow to Opus 4.7 and need help building the underwriting Project structure, the prompt library, and the evaluation framework, The AI Consulting Network specializes in exactly this. For a comparison of Claude against ChatGPT on a specific CRE workflow, see Claude vs ChatGPT property valuation.
Limitations Worth Knowing
- The tokenizer change increases real-world cost. Plan for 20 to 35 percent higher bills on the same workload before factoring in caching savings.
- Self-verification adds latency. Multi-step tasks take incrementally longer on Opus 4.7 than on Opus 4.6.
- Vision improvements help most on poor-quality images. If your documents are already clean PDFs, the visual gain is smaller.
- Coding lift matters most for teams that already use Claude for analytical scripting. If you do not write Python or Excel models in Claude, this upgrade is invisible.
Frequently Asked Questions
Q: Should CRE teams upgrade from Opus 4.6 to Opus 4.7?
A: For most teams, yes, because of self-verification and improved vision. The exception is teams running large volumes of fixed, well-tested prompts on clean documents, where the tokenizer cost increase may exceed the value of the new capabilities. Run a 30 day A/B test on your top three workflows before fully migrating.
Q: How much more expensive is Opus 4.7 than Opus 4.6 in practice?
A: The list prices are identical at $5 per million input and $25 per million output. The new tokenizer produces up to 35 percent more tokens for the same input, so plan on a real-world bill increase of 15 to 25 percent on the same workload after factoring in prompt caching.
Q: Does Opus 4.7 still have the 1 million token context window?
A: Yes. Opus 4.7 keeps the 1 million token context window and 128K output token cap. The combination is what makes Claude particularly strong for CRE work where you load full deal packages, leases, T12s, and prior memos into a single context.
Q: What does task budgets let me do that I could not do before?
A: Task budgets let you cap the total API spend on a single agentic run before the run starts. For CRE, this matters most on overnight portfolio screens or weekly deal review runs where the workload size is hard to predict. Without task budgets, an unbounded run could blow through a daily API budget. With task budgets, the model prioritizes work and finishes inside the cap.