Veyu Lee

Find early users without leaving your coding agent | Building LeadCue

[Veyu Lee @veyulee]

Solo technical founders already build inside Claude Code and Codex.

LeadCue lets them find early users there too.

Give it your product website, docs, or a short description. LeadCue understands the problem you solve, then finds people already showing that need.

Build your product.
Find your early users.
Same place.

[Veyu Lee @veyulee]

@thsottiaux Why? What does “unlimited” mean then?

[Veyu Lee @veyulee]

@thsottiaux That's right! I really love dots!🥰

[Veyu Lee @veyulee]

@thsottiaux Hey, Tibo!
Could we get more RAM for dot’s cloud computer? Mine keeps hitting browser crashes and timeouts during coding and browser work.

[Veyu Lee @veyulee]

@thsottiaux I encountered a serious bug: the dot always gets interrupted during long tasks, doesn't reply to messages, and only resuming work after a phone call.

[Veyu Lee @veyulee]

@thsottiaux @CtrlAltDwayne Tibo, I really want to upgrade to 20xpro. Can you solve this problem first 😭

[Veyu Lee @veyulee]

@ns123abc The information you provided doesn't seem to prove anything. Can you provide more valid evidence?

[Veyu Lee @veyulee]

@kimmonismus 实际上Astra也会写很多，只是写成了文件。不管是为了后续轮次工作的准确性或是专业人士的审查，模型当然需要输出比较多的内容。
所以只能说OpenAI的harness拥有非常好的用户体验，这也是我更喜欢openai的原因之一。

[Veyu Lee @veyulee]

Astra’s pretraining gets all the hype, but its post-training may be even more impressive. Reasoning effort scales hard: High → XHigh can feel like a different model on complex open-ended problems. Public benchmarks barely show it. Internal evals are the real signal.

[OpenAI @OpenAI]

This is GPT-6 Astra.

Anything you can do on a computer, Astra can do for you. Fast.

[Veyu Lee @veyulee]

@djcows I would buy it😍

[Veyu Lee @veyulee]

@thsottiaux Astra, thanks

[Veyu Lee @veyulee]

@pauline_makes I guess kids are gonna love this.

[Veyu Lee @veyulee]

A new AI startup playbook: optimize a specific part of open-source models and inference infrastructure, then offer faster performance at a lower cost.

[Hannah Chung @hannah_chuu]

@OneTriangleAI (YC S26) is building the fastest lightweight, cheap inference and hosting @deepseek_ai V4 Flash today at $0.15/M input and $0.35/M output tokens. Live at https://t.co/3xk6ys5T3W.

Our group of @MIT grads and olympiad medalists from @GoogleDeepMind, @JaneStreetGroup, and @NeurIPSConf /ICML publishers are setting a new standard for cost and latency efficiency using KV cache optimizations.

[Veyu Lee @veyulee]

I think a major source of competitive advantage going forward will be how quickly people and organizations adapt to new models.

As work becomes increasingly agent-driven and major models ship almost every week, workflows can’t stay static.

Every model feels different—even upgrades within the same model family can require a different way of prompting, delegating, and reviewing work.

From what I’ve seen, someone who really knows how to use a model can be far more productive than someone with 10x the quota but little idea how to work with it.

The scarce resource may not be tokens.

It may be adaptability.

[Veyu Lee @veyulee]

@SPAC89 Why do I feel like reading code is okay, but once it comes to writing code, it's easy to miss related code changes?

[Veyu Lee @veyulee]

I think we need Luna Ultrafast

[OpenAI @OpenAI]

Previewing Ultrafast mode: GPT-5.6 Sol at up to 14x the speed.

Launching first in the OpenAI API to a select group of customers with expanded access to more businesses as capacity grows.

[Veyu Lee @veyulee]

@sahill_og "help me commit" save my whole day

[Veyu Lee @veyulee]

@thsottiaux Is this a hint?

[Veyu Lee @veyulee]

We changed how LeadCue charges.

Early-stage founders shouldn’t have to wonder whether another lead search is worth hitting a paywall.

So we removed the paywall from finding leads.

You can now use LeadCue to research and find potential customers whenever you want, for free.

The free plan lets you maintain up to 30 saved leads across all of your ICP lists.

When you reach 30, you can simply delete leads you no longer need and keep going — or unlock unlimited saved leads for $9/month.

The idea is simple:

LeadCue should be useful before you have customers, not become another cost you need customers to justify.

[Veyu Lee @veyulee]

Model cost was a barrier to running LeadCue continuously. That’s now largely gone.

I tested Luna xhigh/max for lead discovery and CMO Mode. Quality was close to Sol medium, at a fraction of the cost.

Daily user discovery and outreach are now far cheaper.

[OpenAI @OpenAI]

We are committed to pushing the model frontier across cost efficiency, capability, and speed.

Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% , and offering a faster option for GPT-5.6 Sol in the API.

Luna and Terra’s lower prices are reflected in how usage is counted in Codex and ChatGPT Work, so your usage goes further.