The headline changes in this release are related to the Python plugin host:
- Upgraded Python 3.8 to Python 3.14
- Python 3.3 is now disabled by default
By default, Sublime plugins now all run under the Python 3.14. The 3.8 plugin host was removed (Python 3.8 reached EOL in Oct 2024), and the 3.3 plugin host is deprecated, and will be removed in a future release.
This is great for plugin devs. They get improvements to syntax, performance, type hints, the standard library, etc, and they can depend on modern versions of Python packages. Many packages no longer receive updates for 3.8 (e.g. https://github.com/psf/requests), let alone 3.3, and many others simply don't run on these older python versions (e.g. https://github.com/xberg-io/tree-sitter-language-pack).
Like the plugin systems for Emacs/Neovim/VS Code (and unlike the sandboxed systems for e.g. Zed and Helix), Sublime's plugin API has low-level access to the editor, and the runtime has full OS access.
I think Sublime was smart to bet on Python for the plugin system. They released the plugin system in 2009, before Python was everywhere. The super flexible plugin API and the huge package ecosystem is a powerful combo.
Emacs doesn't have "plugins", they are called "packages" for a reason. Every Emacs package is not just a plug-n-play "product", it's a recipe, you can use them as libraries, cherry-picking behaviors you want without ever executing the main features of the package.
That is not a good way to think about this. I don't have deep knowledge of how LLM's work, but the following is accurate enough to illustrate the point.
Let's say the LLM is in the middle of text generation and "decides" that the next token is "dog" with p=0.55, or "cat" with p=0.45. With a temperature of 0, the model always picks dog, because it's the most likely next token. With a temperature of 1 the model picks dog 55% of the time and pick cat 45% of the time.
With this watermarking scheme, the model might alter these probabilities s.t. p_dog for this particular generated token goes up or down. Let's say it does down, s.t. p_dog is now 0.45 and p_cat=0.55. Now, with T=1 the model picks cat 55% of the time and dog 45% of the time. Regardless of whether the "watermarking function" raises or lowers p_dog, the probability distribution for this token has changed, and whatever math this trillion dollar company and its brainiacs came up with to decide that p_dog ought to be 0.55 has been "adulterated". As others have mentioned there is no way around this.
---
Regarding the watermarking scheme, it works because it doesn't just alter p_dog for this single output token. It alters probabilities for many of the generated tokens (it could do this to all of the output tokens; it's an implementation detail). E.g. at token N, it favors "cat", at token N+1 it favors "house", etc. This way, if you have the secret key that lets you generate the watermarking function for any output token, you can analyze a run of tokens and check whether it's likely they were generated according to your watermarking scheme. The longer the run of tokens, the more certain this check becomes (it becomes extremely certain quite fast).
> In a bold move that signals a major shift in Mexico’s approach to environmental protection, the Federal Attorney’s Office for Environmental Protection (Profepa) has launched an aggressive campaign against unauthorized real estate developments along the picturesque Yucatán coast.
I would take this story with a huge grain of salt. There's illegal development everywhere in Mexico. Here's a story (one of many, and from an actual newspaper) about illegal development in Tepoztlán: https://www.jornada.com.mx/noticia/2022/08/15/estados/edific.... This goes on in the open -- you can look towards the mountains from town and see fancy houses that are literally within the national park.
I picked Tepoztlán because it's very beautiful, very close to the capital, and is one of Mexico's most visited "Pueblos Mágicos" (the most visited so far in 2026). If the government isn't stopping obvious illegal development here then one would wager they aren't really stopping it anywhere. Insitutions are weak, so crackdowns are limited and temporary, and they're usually just aimed at developers that aren't properly "connected". "For my friends, everything; for my enemies, the law" and all that.
It isn't unusual for an alcoholic to quit drinking for real. But the rate of failing is high enough that unless they are getting serious help in quitting you should assume failure.
- Upgraded Python 3.8 to Python 3.14
- Python 3.3 is now disabled by default
By default, Sublime plugins now all run under the Python 3.14. The 3.8 plugin host was removed (Python 3.8 reached EOL in Oct 2024), and the 3.3 plugin host is deprecated, and will be removed in a future release.
This is great for plugin devs. They get improvements to syntax, performance, type hints, the standard library, etc, and they can depend on modern versions of Python packages. Many packages no longer receive updates for 3.8 (e.g. https://github.com/psf/requests), let alone 3.3, and many others simply don't run on these older python versions (e.g. https://github.com/xberg-io/tree-sitter-language-pack).
Like the plugin systems for Emacs/Neovim/VS Code (and unlike the sandboxed systems for e.g. Zed and Helix), Sublime's plugin API has low-level access to the editor, and the runtime has full OS access.
I think Sublime was smart to bet on Python for the plugin system. They released the plugin system in 2009, before Python was everywhere. The super flexible plugin API and the huge package ecosystem is a powerful combo.