Cortega Legal

High-quality legal work.
Deep research.
Answers you can trust.

Research the matter, challenge the reasoning and check the sources. Cortega combines models for drafting, review and verification, with the evidence and unresolved questions alongside the answer.

A lawyer examining and annotating documents
91.4%Criteria met$0.24Token cost per task

1,743 public Harvey LAB tasks
October 2026 · Graded by Cortega
Token cost excludes verification.

Results, comparisons and limits →

Quality built into governance

More than one model’s first answer.

Cortega’s inline AI quality boosters combine research, independent drafting, review and source checks. Its mixture-of-models approach separates responsibilities, following the division of work in a law firm: research, drafting and review before the lawyer’s decision.

  1. 01 / Research

    Build the evidence

    Read the matter documents and use connected research sources. Keep the passages behind each point.

  2. 02 / Draft

    Compare approaches

    Where enabled, independent drafts surface different points. A merge brings the supported work together.

  3. 03 / Review

    Challenge the answer

    A separate review checks for omissions and incomplete work. Models can take different roles through Cortega’s governed routes.

  4. 04 / Verify

    Inspect the support

    Check document quotations and available citation records. Show missing evidence and verification status for the lawyer’s review.

Explanatory workflow. Checks and research sources depend on the tool and configuration; a source match does not establish a legal conclusion.

What makes the research deep?

The workflow reads the underlying file, follows relevant sources through connected research tools, compares competing positions and reviews what the draft leaves out. Research access depends on the practice’s connectors. Unavailable sources and unresolved authority checks remain visible.

Measured quality / Model choice

Why we benchmarked with GLM Flash.

We chose GLM-5.3 Flash as the primary model to evaluate useful legal work at a low token cost. Cortega’s October public Harvey LAB run met 91.4% of criteria across 1,743 tasks, at $0.24 per task in tokens.

Quality comes from the configured workflow as well as the underlying model. Open-weight models also preserve a choice of private hosting, including infrastructure inside your network.

See the model comparisons and method →
October 2026 / Cortega evaluation91.4%

of rubric criteria met


GLM Flash as primary model

Collaborative workflow · $0.24 token cost per task

Read results and limits →

Graded by Cortega, before the separate citation check. Token cost excludes verification. Criteria met is not the share of complete tasks.

How it works

Research, review and test the work.

Start with the motion, witness or document in front of you.

Review arguments

Write the opposition to a motion. Check the argument against the record, then work through the weaknesses and what is still unknown.

Stress test · Motion test · Expert pressure test · Suppression test

Explore this workflow →
Motion review from the current application using invented sample documents.
Motion review from the current application using invented sample documents. Select to enlarge.
How to read the sample results

These are actual application views using invented sample documents. Results are research aids for a lawyer to review. Authorities are labelled with their verification status. Document quotation matches do not establish that a legal conclusion is correct.

What the review gives you

Check the evidence behind the answer.

↗

Where the position is weakest

Ranked points to work through, with the other side’s argument and a possible answer.

“

The words behind the point

The source quotation and document line, so the lawyer can examine the record.

?

What still needs checking

Missing facts, unread material and unresolved points remain visible.

A legal team reviewing a matter together

Dedicated practice environment

A separate environment for each practice.

Each practice has its own web address and standalone environment. Practice data is not shared with other practices.

  • Private models inside your network, foundation models, or both.
  • Your own provider keys.
  • On premises, your AWS account or rented GPUs.
Explore deployment →

Practice context

Different files.
Different points of attack.

These document tools have narrower scope. Each practice page states the relevant rule and statute gaps.

Also in the workspace: matter documents, Word redlines, citation checking and privilege review →

Questions

Before you try it.

What makes an answer trustworthy?

Inspect the source quotation, document line, citation-check status and unanswered questions. Cortega combines drafting and review with these checks. The lawyer still assesses the facts, current law and conclusion before using the answer.

Which courts and rules are covered?

Civil deadlines cover federal, D.C., Maryland and Virginia cases. The rule-based tools identify the text they read and their gaps. Coverage varies by tool: selected Maryland discovery rules are included, while Maryland motion, expert and appeal packs have gaps. Each practice page lists the source status.

Can the practice use its own models?

Yes. Use private models inside the practice network, foundation models, the practice’s provider keys or a combination.

Does it decide what to argue or file?

No. Results are research aids. The lawyer checks the facts, authorities and conclusions and decides what to argue or file.

Cortega Legal

Review a sample matter with us.

Book a demo →