Lex V2: Personas
Hi there! Nathan here, founder of Lex (https://lex.page), the modern writing platform with powerful AI editing tools built in.
As we mentioned in the last couple of emails, over the next several weeks we'll be rolling out a series of five major releases that will significantly improve Lex. So much so, that we’ve come to think of them as collectively forming Lex V2.
Today, the third of those five releases:
Introducing Personas
At first, we weren’t crazy about this idea.
Creating “personalities” for the AI in Lex seemed like just a gimmick. But then we saw a semi-viral persona, tried it, and it only took a couple of messages to completely shift our perspective.
We realized that many of our frustrations with AI weren’t deep attributes of the model, and can be fixed surprisingly easily with the right system prompt. It’s hard to trust an overly cheerful, generic AI voice that always seems to agree with you. But as it turns out, AI doesn’t always have to be that way.
Here’s a feel for the difference, using Vector—one of our new default personas.
While we personally love Vector, we built a few other options if you want something different. Test them out yourself: head to lex.page, open Chat, and click “Chatting with” in the top left corner.
You also can build your own personas, customizing Lex to be the ideal writing partner for you. If you want to learn more, check out this guide.
We tuned up our standard Lex persona too—focusing it more on clear, actionable feedback, concise edits, and resisting the habit to just tell you what it thinks you want to hear.
💡 Pro tip: personas are available to all users, but they only really shine with top shelf AI. Candidly, free-tier LLMs often miss the nuance that make a persona click.
If you want to give personas a fair shot, we recommend signing up for a free week of Lex Pro and using them with premium models like GPT-4.1 or Claude 3.5 Sonnet.
In other news…
More control over your AI editor
You know that frustrating moment when you want high level feedback and get line-by-line edits instead? Or when you're looking for specific rewrites but get vague suggestions?
We fixed that. Now you can control exactly how Lex responds:
By default, ‘auto’ is pretty good at figuring out the kind of help you need in the moment—but you can also toggle to ‘edit’ or ‘chat’ when you want more control.
A flood of new AI models in Lex
A list of the recently released models that we’re adding to Lex Pro—and an update on our default Pro and free models!
Premium models added to Lex Pro.
GPT-4.1 — Exceptional at following nuanced instructions and handling complex context. Massive 1M+ token window. We were so impressed we swapped it with Claude 3.5 Sonnet as our default Lex Pro LLM!
o3 — OpenAI’s latest cost-effective “reasoning” model. Faster than older models, handles complex logic well, but can be a little rigid with creative tasks. Solid for fast edits, explanations, and when you need lots of rapid answers.
o4 mini — Optimized for speed. Works well for basic edits, grammar checks, and quick brainstorming. Not as nuanced as larger models, but perfect when you need immediate feedback on more straightforward tasks.
Gemini 2.5 Pro — Excels at complex analysis and technical editing. Handles straightforward requests well, has a huge context window (1M+ tokens), but unpredictable for creative writing—brilliant sometimes, but other times oddly off.
Grok 3 — Strong on technical analysis and direct feedback. 128K token context window (with recent AI advances, this is now on the smaller side for premium models—wild right?). Solid for technical writers who need straight-to-the-point responses.
Added to Lex’s free tier:
GPT-4.1 mini — Streamlined version of 4.1. Optimized for speed and cost. Strong performance on basic writing tasks, but trades nuance for faster response times. Great for quick edits and drafting. We swapped it with Claude 3.5 Haiku as our default Lex free LLM!
That's release #3 of Lex V2! Two more big ones coming in the next few weeks. As always, we'd love to hear your thoughts—just reply to this email.
—Nathan