Harvey Release Notes

Gemini 3 Pro in public preview

Google launched their latest frontier model, Gemini 3 Pro, in public preview and we’re sharing results from Harvey’s early testing.

Gemini 3 Pro is the latest model in the Gemini family and represents a meaningful step forward over prior snapshots. During early access evaluations, the Harvey team observed stronger reasoning, cleaner structure, and more consistent style than prior Gemini models across our BigLaw Bench benchmark, product-focused evaluations, and unstructured lawyer usage. Below, we share a look into these results.

BigLaw Bench Performance

Gemini 3 Pro scored 87.9% on BigLaw Bench, an improvement over Gemini 2.5 Pro (85%) and in line with other frontier models. It particularly excelled at transactional work and in the litigation drafting sub-tasks, with better tone control, stylistic polish, and creative enhancements to content when appropriate.

Product‑Focused Evaluations

In product‑oriented tests, Gemini 3 Pro was structured, stylistically consistent, and well‑formatted. It follows instructions tightly — including fine‑grained style requirements — and produces outputs with clear headers and correctly rendered tables. The model also performs especially well on long, multi‑item tasks, demonstrating robust reasoning over extended context.

In‑the‑Wild Lawyer Usage

Unstructured usage by our Applied Legal Research team points to meaningful progress in legal reasoning, drafting quality, and formatting discipline. Outputs are easy to navigate, with clear sections and logical flow. The model shows high instruction‑following across tone, formatting, and style, and adapts well to different drafting contexts.

“Gemini 3 Pro advances on three fronts that matter to legal work: long‑context reasoning, structured drafting, and tone control. It lifts transactional drafting and shows clear gains in litigation writing while keeping outputs well‑organized and on‑brand. We’re thrilled to be bringing this model to our clients in Harvey soon.”

Niko Grupen
Head of Applied Research

We are working in parallel with the teams at Google and DeepMind to bring the model towards General Availability, at which point the model will be made available within Harvey. Stay tuned for more details!

Original source Report a problem

Announcing GPT-5.1 is now available in Harvey.

Today we are making GPT-5.1 available in Harvey.

GPT-5.1 is OpenAI’s newest frontier model, which was trained with an emphasis on dynamic thinking times as well as improvements in formatting, style, and personality. Early access evaluations point to GPT-5.1 providing more detailed and better structured outputs than its predecessor. We expect GPT-5.1 to deliver more complete and structured work products in research, strategic analysis, and drafting.

We’ve also observed improvements in instruction following, which makes GPT-5.1 particularly well-suited for workflows where models need to understand our clients’ exacting standards and their unique style. On our BigLaw Bench evaluation suite, GPT-5.1 achieved the highest score yet for the GPT family of models at 91.8%. This underscores its potential to handle the rigor and nuance of elite legal practice.

“Model performance is improving rapidly. GPT-5.1 shows notable improvements in legal reasoning and instruction-following, which is significant progress in just a few short months since GPT-5's release. With this launch, Harvey clients in professional services are once again at the frontier of model capability.”

Niko Grupen
Head of Applied Research

We’ll be rolling out GPT-5.1 in the model selector to eligible clients in the US and EU over the next 24 hours.

Original source Report a problem

GPT-5.1 is available in Harvey’s Model Selector across Assistant, Vault, and Workflow Builder.

Release Date

Nov 13, 2025

Tier

Tier 3

Categories

Model Updates

Availability

All regions except AU

What's New

We’ve added GPT-5.1 to Harvey, offering you access to OpenAI's latest model. With this release you can expect:

• Improved legal reasoning: Scored 91.8% on Harvey’s BigLaw Bench evaluation.
• More structured drafting: Produces clearer, more consistent outputs.
• Better instruction following: Adapts to firm- or client-specific style and tone.
• Dynamic reasoning: Adjusts response depth for complex prompts.
• More reliable tool use: Handles multi-step research and drafting tasks with fewer errors.
  Learn more here in our blog post.

Why It Matters

GPT-5.1 delivers more accurate and complete outputs for legal research, document comparison, and drafting. Users can expect fewer edits, more consistent formatting, and improved performance in multi-step workflows.

How to Use

• Model Selector: Choose GPT-5.1 in the selector within Assistant, Vault, or Workflow Builder.
• Access control: Model Selector is managed through Access Control & Permission Management.
• Rollout: Enabled for US and EU workspaces over the next 24 hours.
• AU support: Coming soon.
  If you don’t see the Model Selector, contact your workspace admin or Customer Success Manager.

FAQs

Q: How is GPT-5.1 different from GPT-5?
A: It provides more structured, accurate outputs and stronger instruction following.

Q: Will GPT-5 still be available?
A: Yes. Both GPT-5 and GPT-5.1 will remain available for now.

Q: Where can I use GPT-5.1?
A: In Assistant, Vault, and Workflow Builder, via the Model Selector.

Q: Who has access?
A: US and EU workspaces; AU access will follow.

Original source Report a problem

Embedded Experience

Use Harvey in your day-to-day tools so you can draft, review, and respond without changing how you work.

Harvey for Outlook Add-In

Use Harvey’s reasoning and drafting capabilities directly within your email threads to analyze content, draft replies, and store important materials in Vault.

Harvey in iManage

Work seamlessly between iManage and Harvey by selecting documents and asking Harvey questions without leaving iManage.

Redact Workflow in Harvey for Word Add-In

Easily remove sensitive information from Word documents using the new Redact workflow, now including metadata redaction.

Fill a Template Workflow in Word

Quickly fill Word templates with uploaded information, and Harvey will populate standard templates automatically to save you time on document creation.

Convert to a Template Workflow in Word

Turn any document into a reusable template. Harvey will automatically detect names, addresses, and other variable fields, replacing them with placeholders.

Comments in Word Documents

Get answers that account for the full conversation around your document, not just its text. Harvey now can read and understand comments in Word.

Playbook Review History

View and resume your past playbook reviews on a specific document. Track which playbooks you've run and easily return to previous analyses.

Highlight Reviewed Clauses

Highlight clauses in the contract that were reviewed by the playbook. Harvey will leave comments explaining which rule each clause is associated with.

More Context, Deeper Insights

Get better answers to complex legal questions by drawing on more sources, stronger reasoning, and richer context.

Ask LexisNexis®

Answer questions, draft memos, and perform follow-ups with trusted, AI-powered insights from LexisNexis® primary law.

Deep Research with Any Source

Synthesizes across multiple trusted sources — like Vault, iManage, web, file uploads and local legal data sources — to deliver comprehensive, well-cited reports that help you understand the full picture faster.

Multi-Source Query

Query across two sources at once, giving you broader access to the context you need without switching tools or re-uploading files — or switch between different knowledge sources within a thread.

Agentic Search

When querying over uploaded documents or Vault projects, Harvey will intelligently perform iterative searches until it has the information it needs to produce the best possible response to the query.

GPT-5 in Assistant

GPT-5 is now the default model for all chat and draft responses, bringing major upgrades in reasoning, reliability, and writing quality.

Managing Embedded Files in Custom Workflows

Quickly add, delete, and replace all of the embedded files in a custom workflow with a new file management UI.

Local and Vault Files in Workflows

Quickly run Harvey and custom workflows that combine local file uploads, DMS files, and Vault files in a single session, so you don’t have to split up different file types.

Web in Custom Workflows

Ground workflows in up-to-date, real-world information by embedding web search into custom workflows.

Localized Knowledge Sources: Canada, Germany, and Austria

Enable precise, jurisdiction-specific answers directly in Assistant with new legal knowledge sources for Austria (RIS), Germany (Gesetze-im-Internet), and Canada (Justice Laws & Ontario e-Laws).

Improved Governance Controls

More easily manage access, organize work, and keep projects running smoothly.

Admin Feature Controls

Easily enable new features across your team with simplified admin controls and dedicated pages for each opt-in feature.

Flexible Query Organization

Organize projects and share findings easily by choosing which project to save each query in.

Transfer Ownership of Vault Projects

Easily transfer project ownership when responsibilities change — keeping projects organized and collaboration seamless.

User Groups

Organize workspace members with public and private groups, making it easier to manage sharing, personalize collaboration, and control access across matters and departments.

Simplified Experience

Small-but-mighty updates that make Harvey easier to navigate, more intuitive to use and better aligned with how you work.

Knowledge Source UI Improvements

Select and refine knowledge sources in Assistant with greater clarity. Source details now highlight descriptions and data source information in both the source picklist and filter chips.

Improved Sources Experience

A redesigned sources panel makes it easier to navigate, search, and interpret citations — with clearer icons, a new search bar, and auto-updating sources that stay aligned with your current thread view.

Polish and Bug Fixes

Every detail matters when you’re doing critical work in Harvey, so we’ve fixed and improved dozens of small details. We’re also refreshing and polishing different areas of the product to make it easier and more enjoyable to use, including updated source chips and icons across Assistant and Vault.

That’s all for November. See you next month!

Original source Report a problem

Harvey’s platform is built for enterprise-grade security, and new admin control features make governance easier than ever.

As your organization scales, maintaining visibility and control over who can access what becomes complex. To better address this, we’re releasing new features to help teams scale securely while maintaining the governance their work requires.

Whether they’re configuring permissions, structuring groups, or syncing their identity management system, finer admin controls give teams the ability to keep their entire organization moving efficiently, securely, and confidently on a platform that meets the most stringent industry security standards.

Centralize Control

We've introduced a new settings experience that lets admins discover and manage all workspace features in one place. Admins can now enable features like integrations, models, and workflows — and immediately control which users and roles have access — directly from each feature's dedicated settings page, eliminating the need to hunt through separate roles and permissions screens.

Additionally, new user groups make it easy to mirror permissions to the way your organization actually works. Create groups for different practice groups or for regional offices and client teams, and configure permissions across them instantly. This helps users set up teams faster, centralize visibility into access settings, and gain complete confidence that their team’s environment is always secure and aligned.

When a new professional joins a group, their permissions automatically match their peers’. When someone moves to a new team, their access updates in real time. This leads to fewer onboarding bottlenecks and access errors, even in fast-moving organizations.

Teams can also receive and share resources, allowing users to distribute workflows and templates securely across groups. That means every team can operate at the same high standard, while irrelevant or sensitive information stays exactly where it should.

These simple changes help administrators easily manage firm resources without compromising control.

Automatically Sync Security

Security is never optional in legal work — it’s the foundation of every client relationship and every matter handled. The vast majority of law firms already use identity providers like Entra AD and Duo to manage user permissions. To support these firms, Harvey will introduce full support for System for Cross-Domain Identity Management (SCIM) later this year. That means teams will be able to connect Harvey directly to the systems they already use to manage access.

With SCIM support, account provisioning takes place automatically and user groups are synced to an organization’s system of record. This ensures new users get the right Harvey access immediately, and that permissions are up-to-date. Additionally, SCIM reduces manual setup, making it easier for administrators to save time and maintain proper access.

Maintain Enterprise-Grade Security With Ease

With these latest updates, Harvey makes it even easier for organizations to configure permissions and manage access, helping them confidently ensure security at scale. Admin controls are available now to all Harvey customers from within your account, and SCIM support enters early access later this year. These features are part of our ongoing commitment to building Harvey with trust as a core pillar, not an afterthought.

If you’re not a customer yet and want to learn more about how Harvey can help your organization, reach out for a demo.

Original source Report a problem

Harvey now supports querying across two sources at once — giving you broader access to the context you need without switching tools or re-uploading files.

What's New

Now you can combine two sources (for example, Vault + Web or Docs + EDGAR) in a single Assistant query. This allows Harvey to pull information from different contexts — such as precedent documents, tax databases, or public filings — to produce richer, more accurate answers. You can now:

• Ask more complex questions that span two different sources.
• Generate more comprehensive, accurate responses by combining all relevant context.
• Save time by avoiding the need to manually consolidate documents or move between sources.

Deep Research vs Multiple Sources

Think of Multi-Source and Deep Research as complementary features:

Feature | What it Affects | Primary Benefit | Output Style
Multi-source query | Breadth — how many sources Harvey consults | Broader, multi-context reasoning | Flexible outputs (tables, lists, drafts)
Deep Research | Depth — how deeply Harvey explores each source | Exhaustive research & synthesis | Detailed, report-style summaries

Why it Matters

Previously, queries were limited to a single source—requiring you to jump between contexts or repeat searches. Now, you can bring multiple information streams together in one place, improving efficiency and accuracy across use cases like research, diligence, and drafting.

How to Use

Querying across two sources follows the same process as a standard Assistant query:

• Enter your prompt in the text field.
• Click Files + sources.
• Select two sources (for example, Vault + Web, Docs + EDGAR, or UK Tax + Web).
• Run your query as usual.

Harvey will automatically determine how to retrieve and combine results from both sources.

Prompt Techniques

We strongly recommend including specific language in your prompt instructing Harvey how to use each source instead of issuing general instructions.

Example Prompts

Sources | Poor Prompt | Better Prompt
Docs + Web | “Are there any blog posts about this?” | “Search the web for blog posts about the attached case.”
Web + EDGAR | “See if there’s any ongoing litigation for Plug Power Inc. and then find any negative press.” | “Act as a corporate attorney conducting due diligence for a potential acquisition of Plug Power Inc. (NASDAQ: PLUG). Identify any ongoing litigation in 10-Ks or 8-Ks on EDGAR, then search the web for related public litigation or negative press.”
UK Tax + Web | “What taxes apply to the purchase of land by an individual and how can I minimize those taxes?” | “Find which UK taxes apply to the purchase of land by an individual, then search the web for guidance on how to minimize those taxes.”
Sweden Case Law + Docs | “Does this accurately describe the laws on pollution into navigable waters?” | “Review the attached document’s treatment of Swedish law on pollution into navigable waters, then check Swedish case law to determine if it is accurate.”

Notes on Using a Jurisdiction-specific Source
When using a jurisdiction-specific source, include the jurisdiction in your prompt for best results. For cross-jurisdictional queries (e.g. Colombian and Mexican case law), specify how Harvey should research each jurisdiction.
Example:
• Instead of, "Compare and contrast the laws related to passenger vehicle safety standards," try, "Compare and contrast the laws in both Colombia and Mexico related to passenger vehicle safety standards."

Notes and Limitations
• As you work in a thread, you can switch sources (up to two) to apply whichever sources are most relevant to your follow-up queries.
• Multi-source queries are not yet available with:
◦ LexisNexis®
◦ Review Tables

FAQs

Q: Which sources can be combined?
You can combine any two supported sources, including internal document repositories, connected data sources, and web search.

Q: Will this affect existing workflows?
No changes are required. Your existing queries will continue to function as before; Multi-source Querying simply adds new flexibility for more complex research needs.

Original source Report a problem

Harvey’s Deep Research connects across multiple trusted sources—Vault, iManage, web, file uploads and Knowledge Sources —to deliver comprehensive, well-cited reports that help you understand the full picture faster.

Deep Research allows you to analyze, connect, and synthesize information across multiple trusted knowledge sources. Watch how Deep Research works alongside Vault and a regional knowledge source to produce broader, more detailed insights:

[Video Player]

Why it Matters

Deep Research helps you complete complex analysis faster by drawing on your organization’s documents, data, and approved external content in a single, comprehensive report.

How to Use

Follow Using Deep Research for steps, use cases, and FAQs.

Notes and Limitations

To use Deep Research across multiple knowledge sources, ensure that:

• Deep Research is enabled for your workspace.
• You have access permissions for the knowledge sources you intend to include (e.g., Vault, web, DMS integrations).

Original source Report a problem

Harvey just got smarter—refining its approach to find the right information and deliver more accurate answers from your documents and Vault projects.

Harvey now uses a multi-step search process when answering queries about your uploaded documents or Vault projects. Instead of relying on a single search, Harvey can refine and repeat its searches automatically until it finds the information it needs—producing more complete, reliable answers.

• Smarter searches: Harvey can perform several targeted searches in one query, improving accuracy and coverage.
• Better context: Harvey can review and process a broader set of your relevant materials to deliver higher-quality responses.

Why it Matters
This update helps Harvey deliver more accurate answers with fewer gaps or misinterpretations when working with your own materials. It should improve depth and reliability, especially when searching across large or complex document sets.

FAQs
Q: Do I need to enable agentic search manually?
No, this is automatically applied to supported knowledge sources.

Q: Will this change affect query speed?
You might notice slightly longer processing times as Harvey refines its searches, but this ensures more complete and accurate results.

Q: Which knowledge sources are supported?
Agentic search currently applies to uploaded documents and Vault projects for single-source queries, and it’s automatically active for all sources in multi-source queries.

Original source Report a problem

Redesigned sources panel

A redesigned sources panel will make it easier to navigate, search, and interpret citations — with clearer icons, a new search bar, and auto-updating sources that stay aligned with your current thread view.

We’re rolling out a revised sources panel that makes it easier to explore and understand the materials behind Harvey’s answers. With this update, you'll experience:

• Improved design: Refreshed layout improves readability and makes it easier to browse citations.
• Clear source icons: Updated visuals help you instantly identify the type of each cited source.
• In-panel search: Search through the text of cited materials directly within the panel for faster reference.
• Dynamic syncing: Sources automatically update as you scroll through a thread, ensuring the citations you see always correspond to the currently viewed thread.

Why it Matters
With an updated design, improved navigation, and real-time syncing, you can now view, search, and verify citations more efficiently—especially when working across multiple knowledge sources.

Notes and Limitations
This update only effects the visual appearance of the source panel - it does not change how citations are generated.

Original source Report a problem

Organize workspace members with public and private groups

Harvey now supports Groups, a new way to organize workspace members by practice area, department, client matter, or shared interest. Groups make it easier to share resources, control access, and collaborate securely within your workspace.

This update comes with:

• Two Group Types:
  • Private Groups – Used for access control and compliance. Members are added by the group owner or admin.
  • Public Groups – Used for collaboration and knowledge sharing. Members can join or leave freely.
• Group Management Tools:
  • Create, edit, and delete groups directly from Settings ▸ Groups.
  • Filter by group type (public or private) to review membership and access.
  • Manage group details including names, descriptions, and shared resources.
• Group-Based Sharing:
  • Share workflows, playbooks, or Vault projects directly with groups.
  • Choose the appropriate access level — View, Run, or Full Access — when sharing.

Why it Matters

Groups give teams a flexible way to collaborate while maintaining compliance.

• Private Groups ensure that sensitive workflows or client-matter projects stay restricted to authorized users.
• Public Groups help surface relevant resources by practice area, interest, or location — improving discoverability and cross-team engagement.

How to Use

Learn how to Create and Manage Public and Private Groups, including default permissions and FAQs.

Original source Report a problem