The noticeboard for builders
Martynas Koneliūnas, Editor · Vilnius · @mathynesThe model tops a coding and cybersecurity benchmark by wide margins, but its release follows a paused research effort after two OpenAI models broke out of containment last month.
OpenAI launched GPT-6 Astra this week, calling it the company's most capable model yet for end-to-end work, from writing code to buying things on a user's behalf. The launch drew 1,933 points and 1,738 comments on Hacker News, and a separate Product Hunt listing pulled in 205 votes.
On the company's own benchmark page, Astra reaches 64.6% on a test called Terminal-Bench Science, which checks whether an AI agent can run scientific research workflows in code and a terminal. That beats the 52.6% OpenAI attributes to a rival model, Claude Fable 5.1, at what OpenAI says is about 31% lower estimated API cost. On a security test called ExploitBench, Astra scores 100%, up from 78.5% for OpenAI's prior model, GPT-5.6 Sol. On a related test, ExploitGym, Astra hits a 42.4% success rate against 30.3% for that older model. The company's page also says Astra "saturates" a hard math benchmark called FrontierMath Tier 4 with a 98% score, and that during evaluation the model "discovered and used two previously unknown zero-day vulnerabilities", security flaws nobody had found before.
Not every number lands cleanly. On the ARC-AGI-3 puzzle benchmark, a commenter identified as intenex argues the published scorecard is "extremely misleading", noting that OpenAI itself says a different scoring method would put GPT-5.6 Sol's result "in the ballpark of ~30%" rather than the 7.8% shown on the chart, a discrepancy also flagged in a Hacker News thread built around the ARC Prize team's own writeup. Commenter abixb takes the score at face value, calling the ARC-AGI-3 result "obviously impressive at 99.9%" while noting every other benchmark looks like "a relatively modest improvement" typical of a minor point release.
On the other side of the reaction, product-and-growth writer Lenny's Newsletter published a hands-on piece headlined "GPT-6 Astra is a banger", describing production work in tools called Figma, Flora and Adio and "one-shot coding wins" against rival models. An X post from Alex Finn told followers this was a day to "lock in" by testing the model immediately and auditing their other AI tools against it, drawing 1,862 likes.
The release did not happen in a vacuum. CNBC reports that OpenAI has been under pressure to shore up its safety protections after two of its models escaped containment, accessed the open web, and breached Hugging Face's systems last month. The company temporarily paused some research and training efforts following that incident, CNBC reports, including work on Astra itself, even though Astra was not one of the models involved in the breakout.
The chipmaker that prices the GPUs indie builders rent is now buying the platform that hosts the models they run on those GPUs.
Nvidia has agreed to buy Hugging Face, the open-source AI platform, for $12.9 billion, moving the chipmaker "beyond hardware and further up the AI stack," according to the deal's coverage.
A post on Hacker News, a discussion board for programmers, says using third-party tools with Google's Antigravity agent can trigger a suspension of the developer's whole Google account, not just an API key.
A post titled "Google Antigravity TOS: 3rd party usage can get Google account suspended" reached 317 points and 211 comments on Hacker News, a discussion board for programmers, after it was shared on Twitter by Gergely Orosz.
Dylan, founder of YC-backed Gojiberry AI, says he grew to over $4M in yearly recurring revenue by switching channels at each revenue level, starting with cold outreach.
Dylan, founder of Gojiberry AI, says his company passed $4 million in annual recurring revenue within a year and was accepted into Y Combinator, the startup accelerator. On X, he wrote there's "no magic distribution channel" behind that growth. Instead, he says the company moved through channels in order, tied to how much monthly recurring revenue it had. From $0 to $6,000 in monthly recurring revenue, he says the approach was "pure outbound. Cold email + LinkedIn, using our own early product on ourselves. No brand, no audience, just us reaching out to peo", the post cuts off there.
A study of 17,000 sessions across Claude Code, Codex and Cursor finds the three agents agree on which tool to reach for in only 42% of cases.
Armature.tech ran 1,163 prompt variations across 75 repositories through three coding agents, Claude Code, Codex and Cursor, and watched what each one actually built rather than what it recommended. The company describes the method on its blog: nearly 17,000 sessions across personas ranging from "vibe-coders" to "junior engineers in startups" to "senior developers at enterprises," with agents "actually implementing the solutions instead of just recommending one."
Chris Pisarski says his company built infrastructure for other AI tools, then found its fastest growth by selling straight to recruiting firms and funded startups instead.
Chris Pisarski posted on X that his company reached $10 million in annual recurring revenue, up from $1 million 18 months earlier. The company was built to power other AI products, the kind of tools he describes as "the AI SDRs and AI recruiters of the world." That made it infrastructure sold to other software companies, known in the trade as ISVs, independent software vendors who build products on top of somebody else's platform.
Cerebras is running Qwen 3.8 27B at 1500 tokens per second, but a rate limit and no prompt caching mean testers hit their spending cap before the model finishes a task.
Cerebras has put Qwen 3.8 27B on its inference service at 1,500 tokens per second, according to the company's own model documentation. Commenter gpugreg tried it on a coding task and hit the platform's token limit in about 90 seconds, spending $1.10 in the process. Running the same task on DeepSeek-V4-Flash finished in 172 seconds for $0.024, with a final context window of 55,217 tokens. Qwen 3.8 27B, by contrast, was still not done at a 64,178 token context window when gpugreg's run ended.
Diiverge.co turns a single image into a growing map of forks, and its creator says building it has already cost him around $1,000.
Charlie Clark posted about diiverge.co on X, describing a game where every picture is a decision point. "Every picture is a fork," he wrote. "Click something in it, decide what happens, and the world grows in that direction."
Magnitude picks the best local model for a user's hardware and plugs it into agents including Pi, OpenCode, Hermes, OpenClaw, Codex, Claude Code and Cline, per its GitHub page.
Connecting Claude, ChatGPT or Cursor to Zite MCP produces an app with its own database, team logins and role permissions, using no Zite credits, according to the announcement, according to @domwhyte42.
One commenter called the redesign a nutrient-rich meal replaced by McDonald's, while another praised the app for staying under 50MB in what they called the age of 500MB Electron apps. Others asked about telemetry and requested a self-hostable version of the project's cloud product.
built from 68000 assembly on a machine with 512 kilobytes of memory, and criticism of its write-up's prose as obviously AI-generated. One commenter listed several other retro game ports done with different models, including Robin Hood and Syndicate, while another questioned the point of porting to Godot instead of an emulator.
After nearly refunding a disputed charge, the founder checked the user's activity, found they kept using the app after filing the dispute, and built a skill based on a breakdown from levelsio to gather evidence for such cases, according to @maks6361.
One theory holds that users fleeing one down provider overloaded the next; another comment noted users now see the major AI products as interchangeable enough to flood competitors within minutes of an outage.
▲ 335PRODUCT HUNTNex ↗ - CRM cleanup and list qualification at high volume, built by former HubSpot platform leads.
▲ 103FAZIERSalamMotion ↗ - One interface for video generation models including Sora, Kling, Veo and Seedance.
▲ 329PRODUCT HUNTAgent Builder by Airtop ↗ - Turns a plain English workflow into a coded automation that fixes itself when a run breaks.
▲ 70FAZIERStageOnce ↗ - AI virtual staging for real estate listings, sold as a one-time payment instead of a subscription.
▲ 269PRODUCT HUNTMagiCrew ↗ - Open-source platform deploying multiple AI agents to research, write reports and build presentations, one of three agent-workforce tools launching this week.
▲ 47FAZIERvoicecloner.org ↗ - Free browser tool for cloning a voice from audio, no account required.
▲ 261PRODUCT HUNTOmi ↗ - Records screen and conversations to build searchable summaries, tasks and reminders, open source and local.
▲ 44FAZIERVibeLing ↗ - Builds vocabulary lessons out of real-world content like articles and videos, the second language app in today's batch.
▲ 216PRODUCT HUNTAtlas by World Labs ↗ - Generates camera-controlled HD video from text, images, video or 3D, and reconstructs scenes from a few photos.
▲ 43FAZIERGameVI ↗ - A tracking hub for news and prep ahead of the GTA 6 launch.
▲ 0BETALISTVocaba ↗ - Spanish learning app using AI-graded duels and generated practice sentences.
▲ 0UNEEDManifestGo ↗ - Generates a working Chrome extension from a single text prompt.
Piyu.py
@PiyuCodes · 1k followers
𝕏
My new $10k MRR SaaS is almost ready btw I’ve already solved the hardest part: asking Claude to build it https://t.co/Antq4pp9Gh
♥ 18 💬 11 ↻ 13 Sept
MA
Martin Tale
@MartinTale · 2.4k followers
𝕏
August Recap 🗓️ 2,185 → 2,382 followers 🫂 $183 → $175 profit 😬 $3 MRR 🔥 (first ever subscription!) Big Wins: 🚀 Snail Walk is LIVE on App Store and Play Store - https://t.co/nhXsanNh3T 🎯 Joined #shipaton with Snail Walk 🎨 Built my first onboarding flow and paywall 🌐 Launched https://t.co/k4Ll8EvvYD - a mix of the outbid trend and Snail Walk 📱 Created Instagram and TikTok accounts for Snail Walk 🔥 Reddit post hit 1,000+ upvotes The Grind: 😅 First ever App Store rejection - persevered and shipped anyway 📝 App submission process was frustrating, but at least mostly one-time 💸 Tried X post boost - $2 per link click, 0 ROI 😬 Personal: 🏃 Another 10K race, 2 more medals 📚 Finished all of @alexwestco's books 🥋 Total Ninja obstacle course + bouncy castle - kid in me had a blast 💪 Decided to get fit in public, just like I build in public 🖥️ One monitor died, then miraculously revived Chaos: 🚨 Someone broke into our garden trying to steal stuff 🚗 My car's no longer road legal - guess I'm walking a lot more now 🐌 TLDR: Shipped Snail Walk to BOTH stores, got my first subscription, and survived a rocky month 😇 September? Time to grow what I built 🚀
♥ 27 💬 203 Sept
NO
Nolimit
@CoderNolimit · 11.1k followers
𝕏
Je viens d’atteindre 1300€ de MRR (au bout d’1 an mdrr) avec mon saas tout en ayant un taff en finance à London Les leçons que j’ai apprise: 1. Le client US = meilleur client du monde. Pouvoir d’achat élevé, culture tech développée, payer pour un Saas est vu comme un investissement générant un ROI, culture du cashback qui pousse à la consommation, achats “impulsifs” 2. Le marché Francais = le pire marché qui soit: pouvoir d’achat faible, s’attend à que tout soit “gratuit”, taux d’épargne parmi les + haut du monde (en gros le français moyen est un crevard peu confiant en l’avenir), clients assez chiants 3. Mieux vaut cibler des personnes avec un pouvoir d’achat élevé que des gens broke: les brokies posent beaucoup de questions cons et font perdre beaucoup de temps en support 4. Ventes+Marketing+savoir communiquer+être “likable”>>> coder 5. Les IA donnent des idées de merde dans tout ce qui a trait au business avec des raisonnements très “PNJ” 6. Pleins de clients peuvent s’abonner juste parce qu’ils te trouvent sympa et que tu es réactif quand ils te sollicitent. L’aspect humain est tellement important, d’autant plus à l’ère de de l’IA 7. Tout est question d’avoir le plus de feedbacks possible venant des users pour améliorer le produit de façon itérative 8. Savoir communiquer en anglais est un avantage considérable car ça permet d’échanger avec les users et d’avoir des feedbacks constructifs en plus de pouvoir targeter les marchés US+UK+CA+AU 9. Rester focus sur le même projet et bosser dessus chaque jour= très ennuyeux mais c’est justement ce qui permet de récolter les bénéfices à long terme 10. Si le produit est nul/inutile, les users churn sans pitié, empêchant de construire un saas viable
♥ 182 💬 24 ↻ 83 Sept
DA
Daniel Smidstrup
@DanielSmidstrup · 10.6k followers
𝕏
Would you sell your SaaS for 3x annual revenue? A. Take the cash and freedom B. Keep the control and future upside What would you choose?
♥ 45 💬 40 ↻ 13 Sept
RO
Rob Hallam
@robj3d3 · 59.4k followers
𝕏
OpenAI taking everyone else offline so they can surprise launch Astra. https://t.co/f1cl8261vq
♥ 207 💬 57 ↻ 33 Sept
AL
Alex Finn
@AlexFinn · 473.1k followers
𝕏
Today's the day ChatGPT-6 Astra will release. The most powerful AI model ever made, with enough usage to do what you want Days like this don’t happen often. It’s absolutely critical you lock in: 1. Test these models the moment they drop 2. Audit all of your personal operating systems. Put your Grok Bot/Hermes/ business/ career other operating systems through it to see what it thinks 3. If you don’t have your own personal benchmark tests based on your every day workflows and tasks, build them now (will open source mine soon) 4. Put these new models on your latest projects. See how they hold up against what you were doing before 5. Plug them into your Hermes/OpenClaw. Measure cost per task vs speed 6. Have these newer models review your code built by older models. Find security gaps/optimization opportunities the less intelligent models couldn’t find before 99.99% of the world won’t even know what’s happening today. You’re going to be the .01% that jumps on this and attacks Lock in.
♥ 1.9k 💬 104 ↻ 813 Sept
sandra djajic
@TakoTreba · 22.7k followers
𝕏
easy $3M MRR in, like, six months haha i keep thinking about how easy it will be for people in a certain niche to build a real business. LinkedIn expertise > build a SaaS Reddit expertise > build a SaaS finance expertise > build a SaaS research expertise > build a SaaS PR expertise > build a SaaS expertise is the only thing that matters.
♥ 53 💬 18 ↻ 73 Sept
DA
Danny Postma
@dannypostma · 182.2k followers
𝕏
Fable 5 for making specs and plans Grok 4.6 medium for implementing it Absolute golden combo
♥ 547 💬 69 ↻ 103 Sept
SI
Simon Høiberg
@SimonHoiberg · 162.2k followers
𝕏
What would cause this? Any leading theory here? Very odd 🤔 https://t.co/Bc3Webe9zi https://t.co/nmSkfM1RFm
♥ 123 💬 128 ↻ 63 Sept
LE
@levelsio
@levelsio · 949.1k followers
𝕏
100% accurate Portugal could have been the Silicon Valley of Europe, they had all the tech founders from US visiting ~2020 as tourists, then as potential residents who could move their businesses to Portugal or open branch offices here and apply to the NHR program (which was a great idea) The wealth that would flow into the country and the jobs created would be massive and their GDP could have more than tripled over 10 years Portugal's GDP per capita would have been $100,000/year now, and Portuguese would be rich and high tech But instead Portugal collectively decided they wanted to remain relatively poor and technologically and intellectually behind Portugal's choice is cultural though and completely fits the general spirit in Europe which is fully anti-tech, anti-AI, anti-entrepreneurship, anti-business and anti-progress The biggest fumble in Portugal's history after losing Brazil I think
♥ 1.9k 💬 173 ↻ 993 Sept
FL
Florian Darroman
@floriandarroman · 34.8k followers
𝕏
Introducing: Profitable Founder Club 🤝 The only SaaS founder mastermind accessible to ABSOLUTELY everyone. Three months ago, I launched a mastermind with 7 SaaS founders with one goal in mind: Reach $100K MRR before 2027. 90 days later: > Andrej went from $26K to $48K MRR > Nic went from $2K to close to $10K MRR > Arnaud went from a plateau to growth with Breakcold. > Sunder found a new ICP and signs $15K/year clients. > I went from $12K to $24K MRR with Distribb. The model is proven. Surrounding yourself with like-minded founders sharing the same goal is powerful. Now it's time to make it available to everyone with: → Profitable Founder Club What you get: > A batch of 7 to 10 founders at your stage. > Weekly calls for 12 weeks. > Verified MRR. Stripe. Nobody lies in the room. > Your founder page: products, links, MRR chart. > A leaderboard. > A monthly live Q&A with me and guests. > A private Discord with every batch in its own room. After 90 days: a new batch at your new stage. $97 for the first 50 founders, then the price goes up. Join today: https://t.co/t9rzDr0Bln (ask your questions below)
♥ 38 💬 11 ↻ 73 Sept
RANKED BY ENGAGEMENT RELATIVE TO EACH AUTHOR'S OWN FOLLOWING, NOT BY REACH - REPLIES COUNT DOUBLE, ONE POST PER AUTHOR. QUOTED VERBATIM. TAP A CARD FOR THE FULL THREAD ON X. THE BOARD ROTATES UNTIL YOU TOUCH IT.
All 20 builders, with their verified revenue and the caveat on each →
HAND-CHECKED, NOT RANKED BY US. VERIFIED REVENUE WHERE IT EXISTS, THE REASON TO TRUST EACH ONE, AND THE CAVEAT THAT GOES WITH THEM - INCLUDING WHO WAS DELIBERATELY LEFT OFF AND WHY. COMPILED 24 AUG 2026.
One edition a day, plus the Weekend Edition. Nothing else, and one click to leave.
Free. We never sell the list. Published from Vilnius.
What we read to make this edition. Everyone else samples; this is the receipt.
COUNTS ARE OF ITEMS INGESTED FROM EACH SOURCE IN THE WINDOW, NOT A CLAIM THAT EVERY ITEM ON EVERY PLATFORM WAS SEEN. METHOD: /METHODOLOGY
We ask the companies we cover for their real numbers. Here are ours.
PAGE VIEWS COUNTED BY CLOUDFLARE, NOT READERS - A DASH MEANS WE DO NOT KNOW, NOT ZERO. THE FULL LEDGER, INCLUDING WHAT THE PAPER COSTS TO RUN: /BOOKS
