Legal AI vendors are building their own models to cut inference bills and reduce platform dependence

Harvey says it has developed a custom model for legal work called Tenet. Thomson Reuters has said it will soon roll out its own model, called Thomson. Both moves point at the same operational reality: legal AI vendors are trying to stop paying frontier-model tolls on every query and start owning more of the stack themselves.

Bloomberg Law reported the Tenet announcement Tuesday and the Thomson Reuters plan from late July, noting both efforts rely on open-source foundation technology that can be downloaded and customized. That’s a marked change from the last two years, when many legal AI products were essentially “model routers” that flipped between OpenAI, Anthropic, and Google depending on the task.

For enterprise legal operations leaders, this isn’t a branding exercise about whose model is “best.” It’s a contract and governance change. If your vendor runs its own model, pricing mechanics, security ownership, and portability questions all move.

The shift is about unit economics and control, not just model quality

The clearest upside for vendors is cost control. Bloomberg Law cited advisers who expect vendors can improve profitability by directing more queries to their own models instead of paying OpenAI or Anthropic for inference. That matters to buyers because vendor margin pressure has a way of showing up later as usage limits, feature gating, or price resets at renewal.

The operational trade is that responsibility shifts. Bloomberg Law reported ZwillGen AI Division director Brenda Leong’s view that when a company runs its own model, it also takes on security and other obligations that were previously pushed down to the frontier lab. Maintenance and updates don’t disappear, they get reassigned.

In practice, legal departments should read “custom model” as a signal to scrutinize the vendor’s model lifecycle: red-teaming cadence, evaluation sets for hallucination and citation behavior, and how quickly the vendor can roll fixes without breaking workflows. Those are procurement questions, not demo questions.

OpenAI is moving closer to law firms as vendors try to reduce dependency

While vendors like Harvey and Thomson Reuters work to reduce reliance on frontier labs, Business Insider reported OpenAI is moving in the opposite direction: closer to legal buyers. The company hired Jason Boehmig in June, an executive best known for building Ironclad into a major contract lifecycle management player, according to Business Insider.

That hiring move matters because it suggests OpenAI is building legal-specific distribution and product muscle, not just selling APIs. Business Insider also reported OpenAI has been posting roles tied to legal product development, while declining to outline specifics of what it will build.

Bloomberg Law separately reported OpenAI told the outlet it is collaborating with Willkie Farr & Gallagher to deploy ChatGPT Enterprise to every attorney and integrate OpenAI models deeper into the firm’s internal tools. A firmwide deployment is operationally different from a pilot: it forces decisions on identity, logging, matter segmentation, and acceptable use to be resolved at scale.

What legal ops teams should put into 2026 AI specs and renewals

A vendor’s ability to switch among models used to be framed as resilience. Now it can also be a moving target. Bloomberg Law described how legal tech companies have routed tasks across models from OpenAI, Anthropic, and Google; the emerging model-building trend suggests that “what powers the product” may change more frequently, and more quietly, than most legal departments have historically governed.

That lands directly in procurement language. If your contract assumes a named subprocessor (for example, a specific frontier lab), and the vendor swaps to an in-house model, the risk profile changes even if the UI stays the same. It’s also a budgeting issue: vendor-hosted models can change cost curves for heavy users, especially for contract review, discovery-adjacent analysis, and large document summarization where token volume is predictable.

Bloomberg Law reported legal tech advisers expect average attorneys may not notice a change unless model quality shifts. That’s a useful reminder for operators: user satisfaction scores might stay flat while the underlying cost, security posture, and auditability changes materially.

What to ask your vendors and integrators before the next rollout

• Model governance: If the vendor runs a proprietary or customized open-source model (as Bloomberg Law reported for Harvey and Thomson Reuters), what is the update cadence, evaluation methodology, and rollback plan when accuracy regresses on your document types?
• Security and compliance ownership: Which controls are the vendor’s responsibility versus yours now that more of the stack may be vendor-operated, including red-team results, incident notification SLAs, and audit artifacts? Bloomberg Law highlighted that security obligations shift when model maintenance is internal.
• Commercials and cost predictability: How does pricing map to inference consumption? Ask for concrete assumptions: typical token volumes per contract, per brief, per matter, and what happens when usage spikes.
• Portability and competitive overlap: With OpenAI pushing into legal distribution (Business Insider) and deploying ChatGPT Enterprise at Willkie (Bloomberg Law), confirm what happens if a platform provider becomes a direct option for your org. Can you export prompts, matter metadata, and clause libraries in usable formats, and are APIs stable across model changes?