Claude on Jylo: Performance, Cost and Model Selection

What's New

Two new models have been added to the platform as part of version 2.3.13:

• Claude Sonnet 4.6 — a high-performance model that excels at analytical reasoning, interpretation and complex document review.
• Claude Opus 4.6 — Anthropic's most capable model, designed for the most demanding tasks requiring the highest level of comprehension and judgement.

Claude Sonnet 4.5 and Claude Haiku 4.5 are also available to use on Jylo. All Anthropic models are securely hosted on Amazon Web Services and are available in the Europe and United States regions.

The Key Principle: Thinking Versus Following Instructions

If your instruction is precise and well-defined — for example, "extract the rent review date from clause 4.3" or "pull the tenant name and lease start date" — the model does not need to draw on its own reasoning or knowledge. In these cases, the existing OpenAI models such as GPT-4.1 or o3-mini will follow the instruction reliably and give you significantly more processing per credit.

If, however, the task requires the model to interpret, assess or exercise judgement — for instance, evaluating ambiguous break clauses, flagging risk in non-standard lease terms, or providing commentary that goes beyond what is explicitly written in a document — the Claude models deliver meaningfully better results.

Real-World Credit Consumption

To help illustrate the practical impact of model selection, we tested three common Playbooks across four models: GPT-5, GPT-4.1, Claude Sonnet 4.6 and Claude Opus 4.6. The test set comprised five Non-Disclosure Agreements, one Share Purchase Agreement and one Commercial Lease.

Actual consumption will vary with document length and prompt complexity.

GPT-5 and GPT-4.1 are similarly priced overall. GPT-5 offers stronger reasoning whilst GPT-4.1 is faster and can handle larger documents. Claude Sonnet 4.6 sits at roughly 1.5–2x the cost of the OpenAI models, and Claude Opus 4.6 at roughly 2.5–3x. The additional credit consumption reflects the significantly higher reasoning capability of the Claude models — the right choice depends on whether your task benefits from that deeper intelligence.

Recommended Models by Use Case

Reviewing and interrogating single documents

For analytical work on individual leases, reports or risk assessments, Claude Sonnet 4.6 is the recommended choice. It produces noticeably stronger reasoning on complex or ambiguous provisions than GPT4.1 or o3-mini, and is faster thatn GPT5.

Multi-document interrogation

Claude Sonnet 4.6 remains the strong default when working across multiple documents in a single session. For particularly complex or high-stakes matters where you need the highest level of comprehension across a full document set, Claude Opus 4.6 is worth the additional credits — but we would recommend treating it as the exception rather than the rule.

Large-scale document review and bulk data extraction

This is where credit consumption scales most quickly and where model selection has the greatest commercial impact. If you are running well-defined extraction prompts across a large volume of documents — pulling specific, structured data points — the existing models such as GPT-4.1 or o3-mini will handle the work reliably at a fraction of the cost. If you find that quality is falling short on more nuanced or judgement-dependent data points, step up to GPT-5 or Claude Sonnet 4.6 for those specific prompts.

Summary

Getting Started

To enable the Claude models on your Jylo Tenant:

1. Navigate to Admin > Settings > Models.
2. Select the region tab where you wish to connect to AWS (Europe or United States).
3. Tick the desired Claude model to enable it.

For guidance on configuring your Playbook prompts to use a specific model, or for a tailored credit estimate based on your workloads, contact us at support@jylo.ai.