Claude Sonnet 4.6 from Anthropic is now available across 8rney for all users. It replaces Sonnet 4.5 as the default model for research, drafting, and document review workflows.
What changed
Sonnet 4.6 shows measurable improvements across the tasks that matter most in legal work. In our internal evaluations, the model performs stronger on multi-step legal reasoning — holding the thread across a long research query where earlier versions would lose track of the original question. Citation grounding is tighter: the model is less likely to assert a legal proposition without anchoring it to a specific source, and when it does cite, the citation is more precisely matched to the claim being made.
On drafting tasks, Sonnet 4.6 produces better-structured output at first pass. Clause organisation is more logical, section hierarchy is more consistent, and the model is less prone to generating filler language that reads like padding rather than substance. Users working on contract drafting and SLP sections should notice the difference without changing anything about how they work.
Where it shows up
The upgrade applies to all model-driven features in 8rney: the research agent, the document review pipeline, the SLP and Moot drafting modules, the matter chat, and Study Tools in student mode. No configuration is needed — if you were using 8rney before today, you are already on Sonnet 4.6.
For users who have specific model preferences, we continue to support model selection in Settings. Opus 4.6 remains available for the most demanding long-context tasks where its additional reasoning depth is worth the higher latency.
On model selection in legal AI
We evaluate every frontier model release against a consistent set of legal tasks before promoting it to default. The evaluation set includes jurisdiction-specific research queries, contract clause extraction, multi-document comparison, IRAC-structured analysis, and citation verification. A model becomes the default when it performs better than the incumbent across that set — not because it achieves higher scores on generic benchmarks, but because it produces better outputs on the actual work our users do.
Sonnet 4.6 cleared that bar. We will continue evaluating future releases on the same standard.