We are rapidly approaching budget season and planning for contingencies in 2027. Perhaps open-weight models should be part of your fall planning.

Last year, DeepSeek created a scare for the American AI industry when it created a model that cost a fraction to develop relative to its U.S. counterparts. OpenAI alleges DeepSeek used a process known as distillation and leveraged its more expensive AI training.

Guest columnist Ken Crutchfield is founder and CEO of Spring Forward Consulting. A regular contributor to LawSites, he has over 40 years of experience in legal, tax and other industries. Read more about him at the bottom of this post.

The results of distillation are similar to creating a lower-resolution version of an image. The low-fidelity version looks identical as long as you don’t zoom in. The “distilled” image is small enough to send through email, but the recipient doesn’t have the high-fidelity original from your smartphone. The analogy isn’t perfect, but the outputs from a distilled model are similar. They are good, but not the original.

OpenAI has released open-weight versions that can be installed on a personal computer.  They don’t include the training data, but they can be modified and are available under an open-source license. OpenAI is also providing API access that allows developers to use distillation. While these are positive steps, it would be financially disruptive for OpenAI to fully embrace them.

Distributed Computing, Distributed LLMs

In the 1960s, timesharing was a way to access the few mainframe computers in existence. Eventually, some businesses could afford their own. As computing costs plummeted, personal computers arrived on the scene in the early 1980s. The dam broke loose, and there were many computing options. Not all applications required the central mainframe or the latest supercomputer. Nowadays, computing is distributed. Some work happens on a laptop, while other tasks happen on servers or in the cloud.

Similarly, open-weight models will usher in a world where legal and business tasks can be distributed across models and environments. Not everything will require the latest frontier model hosted by Anthropic. More work will be accomplished using specialized models under the control of legal tech vendors or legal organizations themselves. To that end, Thomson Reuters recently announced its own language model.

Open-weight, distillation, and locally deployed models are not the same thing, but collectively they are expanding the options for using AI.  When will law firms rely on an open-weight model behind their firewall or in their private cloud? It might be sooner than we think.

Not Every Task Requires the Latest Model

Complicated legal problems and deep research-style tasks will continue to require comprehensive foundation models. But many legal tasks are narrow in scope and repetitive.

Normalizing contract analytics, labeling clauses, and extracting key terms, dates, and obligations were accomplished using pre-ChatGPT language models like Google’s BERT.

The point is that not every task requires the newest model.  We are rapidly approaching a world where agents may be able to select a local model for specific tasks and only call a frontier model when difficult analysis or unusual language is encountered.

The standard for legal work is typically 100% accuracy. If human review is required, how much should an organization pay for greater model accuracy? And does that premium vary by task and consequence of error? Perhaps an open-weight model is good enough for some applications.

I spoke with a legal tech provider using an open-weight model to write their software. Their development is completed on local machines, so they don’t pay for tokens.  There is no guesswork about token costs.

Legal innovators should begin to ask, “Are there less expensive ways to perform a task reliably, securely, and at the required level of quality?”

I expect the expanding use and options for open-weight models will benefit the legal industry even if it doesn’t benefit the foundation model providers.

Bargaining Power Shift

Open-weight models may also shift commercial power in the legal industry.

Foundation model providers have leverage today, but that declines if legal tech vendors have more choices. (Don’t forget about Google, Microsoft, Amazon, Meta, Apple and others.)

Law departments may perform more work in-house or require that outside counsel use their client-controlled systems. Legal service providers will become more cost-effective alternatives. Larger law firms may have more leverage to build custom solutions rather than depend upon legal tech vendors.

Earlier this year, we were talking about AI coding and open-source software. Now open-weight models create even more options and may democratize more legal work.

Chinese Models As a Disruptor

OpenAI has reportedly raised $122 billion. The Magnificent Seven are investing even more and betting billions on massive data center buildouts.  Nvidia is helping its customers guarantee purchases. American AI is capital-intensive, and investors need their ROI.

Token consumption and subscriptions are key elements to monetization and will remain that way until the business model breaks. But distillation and open-weight models put pressure on the approach and act as yet another accelerant.

Enter China.  They now have competitive frontier models with Kimi, DeepSeek, and Alibaba’s Qwen. Their approach has focused on cost efficiency, and they lean more toward open-weight models, which may not require the same data center investment.

There is debate about the originality of Chinese AI, but regardless, there is pressure on the US approach.  America can try to restrict the models, but that won’t stop other countries from using them. The geopolitics around this will be interesting.

What if the Chinese approach wins? Do LLMs become commoditized faster? The legal industry only recently “discovered” open-source software. Maybe the value shifts back to what is wrapped around the LLM.

The open-weight competition may break the monetization of centrally hosted frontier models. If that happens, American AI providers may be forced to distill their own models and follow an open-weight strategy to remain competitive.

The legal profession will benefit if this happens, as the costs to use AI will come down. The same could hold true for legal tech vendors as they are LLM users too.

Practical Planning Tips

2027 is approaching.  Here are three actions to consider as you plan for the next round of change.

  • Experiment with open-weight models. Propose a sandbox for testing. Identify a few repeatable or costly tasks. Develop some experiments and share the results.
  • Budget for your base case.  Don’t project cost savings in your budget without data. Be conservative until you have facts. Use the experimentation as an opportunity to prove feasibility and savings.
  • Consult your vendors. They are often closer to the frontier models. Ask them what they have done and what they are planning. If you need to sign an NDA, do so.

Start Planning Now

Open-weight models will take on greater mindshare. Apple is pursuing on-device models. There is also a small language model embedded in Windows 11 Copilot+ PCs.

Knowing when and how to leverage open-weight models to augment existing solutions and attorney expertise will be a practical step forward to optimize costs and to make effective use of AI. Not al

AI was used in the creation of this article.

Ken Crutchfield, founder and CEO of Spring Forward Consulting, has over 40 years of experience in legal, tax and other industries. Throughout his career, he has focused on growth, innovation and business transformation. His consulting practice advises investors, legal tech startups, firms, and others.As a strategic thinker who understands markets and creating products to meet customer needs, he has worked in start-ups and large enterprises. He has served in General Management capacities in six businesses.

Ken has a pulse on the trends affecting the market. Whether it was the Internet way back in the 1980s or Generative AI, he understands technology and its impact on business.

Crutchfield started his career as an intern with LexisNexis and has worked at Thomson Reuters, Bloomberg, Dun & Bradstreet, and Wolters Kluwer. Ken has an MBA and holds a B.S. in Electrical Engineering from The Ohio State University.