Last week, 25 companies and organizations, including Nvidia, Microsoft, Meta, and Mistral, sent an open letter to Washington arguing that open-weight AI models deserve protection, not restriction. At Anaconda, we support this stance. In fact, everything the letter argues from a policy standpoint, has been core to our ethos for years.
Who owns the value of your data #
The letter makes its case in civic terms: a strong, open AI ecosystem that diffuses into every sector, versus one dominated by a handful of frontier labs. I’d make the same case in commercial terms, because that’s the tension our customers live with everyday. When a company builds on a closed model, it hands a frontier lab the value of its own data with every API call. This can be the right decision in many cases, but open weight models are the only way a company gets to keep that value for itself, instead of feeding it back to whoever owns the model.
That distinction matters more as AI moves from chat interfaces into the infrastructure of how work gets done. A general-purpose closed model is a reasonable default when the task is general. It stops being the right default once the task is narrow, high-volume, and specific to a domain, which describes most of what enterprises run. Domain-specific open models are winning those jobs on their own merits: better accuracy on the narrow task, materially lower inference cost, and full ownership of the value delivered by that model.
AI transparency is table stakes #
Ask the admin managing software at a bank, a hospital system, or a federal agency what they need before they’ll sign off, and you’ll hear: what’s in the training data, can our team audit its reasoning, and does it ever touch production data before we’ve cleared it? Transparency and explainability are critical in regulated industries and the public sector: if you can’t show the training data and reasoning behind a model’s output, you can’t clear the approval process. Closed models are structurally worse at proving any of that, no matter how capable they are.
AI is being built into the operating layer of the economy. A strong and viable open model ecosystem is what will help keep that infrastructure accountable to the people and companies using it, rather than concentrated in the hands of three or four frontier labs.
What we’re already seeing with Anaconda + Kilo #
We don’t have to make this argument in the abstract. We’re already watching customers act on it. Through Kilo, which recently joined Anaconda, we see this daily: teams running proprietary and open-weight models side by side in production, routing each task to whichever model fits the cost and accuracy bar. Kilo built local model support and model freedom as a first-class feature, and built Kilo Gateway specifically to route across Llama, Mistral, Qwen, and other open-weight models alongside closed ones. That’s a bet that the market wants both open and closed models, the same way it always has in software. The infrastructure that wins tends to be whichever one doesn’t force the choice.
Our Kilo team has been writing about the same shift from the vendor side. Their recent post, “OpenRouter, Opus 5, and the Era of Model Freedom,” argues that walled gardens are giving way to routing and choice, with developers wanting the raw capability of closed frontier APIs alongside the privacy and portability of open weights. Their follow-up, “Two AI Stories, One Enterprise Trust Question,” concludes that neither side is safe to bet an entire stack on. A closed lab can have a bad week you only learn about after the fact, and an open release can get cut off by an export control before a team finishes evaluating it. What protects against both failure modes is the same thing Kilo Gateway is built for: never being locked into one model or one vendor.
Why Anaconda cares #
That instinct, building for optionality instead of picking a side, is also why this isn’t a new conversation for Anaconda. Long before “open-weight” was a policy term, we were the company making sure the open-source Python and data science ecosystem was usable, secure, and governed at enterprise scale. Conda and conda-forge exist because “open” and “well-governed” are not opposites, and the trust layer we’ve built for enterprises sits on top of conda itself. That’s the same case being made in Washington this week, applied one layer down the stack, to the models themselves.
It’s why we’re supportive of the letter’s thesis. People deserve the option to choose, inspect, and own the AI they build.
Learn more about the Anaconda Platform**or request a demo*.*