Week 2's credits are in, with three builders splitting 20,000. AshAmplifies took the top 10k for Games of the Globe, which teaches a tabletop classic in five minutes. Vasily Anoshin took 5k for Tower of Words, and anishsharma took 5k for Human Anatomy, a 3D skeleton you can dissect.
🎒 Netlify Back to School, Day 18 of 31. Twelve builders have deployed 97 teaching apps, and the top of the board has turned into a photo finish. The assignment hasn't changed: build an app that teaches one idea, ship it, come back tomorrow. 🧵
i wish i could isolate that from the rest of traffic and give a solid answer. mostly it costs me my sanity when crawlers go overdrive on some val that's inefficient or expensive to run and that takes down a server or two. they _love_ to iterate through chess positions on the val that does that.
Building a little at a time adds up. Maya shares how she started with no-code tools, found Netlify through a hackathon, and deployed 15 projects while building alongside her full-time job.
Poisoned connection pools can bring your application to a halt.
Find out how this happens and ways to prevent it in our latest blog.
planetscale.com/blog/postgre... Open models like DeepSeek, Kimi, and GLM now run in Agent Runners through OpenRouter. Every site from the test is live: www.netlify.com/blog/one-pro... The DeepSeek result holds up. Its best run resembles a mid-tier closed model in both design and language.
DeepSeek V4 Flash built a full coffee shop site for 1.3 credits in one run of our 11-model test. Claude Opus 5 spent 1,055 credits on one run of the same prompt.
Christopher Lamers
www.linkedin.com/in/lamersc
Kwaku O.
www.linkedin.com/in/kjowusu
Benjamin van Pham
www.linkedin.com/in/bengiaa
Thank you! ❤️ A big thank you to everyone who contributed this week: ❤️
Carlos Urias
www.linkedin.com/in/hicarlosu...
Paul Contreras
www.linkedin.com/in/paul-contre
Sam McClenaghan
www.linkedin.com/in/sam-mccle... 202 successful reproductions and counting.
We’re halfway through the Zed Summer Repro Program, with 15 contributors testing reported bugs across their own setups.
Congratulations to our week-three raffle winner, Paul Contreras! 🥳
www.linkedin.com/in/paul-contre Thanks for the flag. Looks like this one is already down, but please keep sending them our way. fraud@netlify.com or the "Report abuse or phishing" option at netlify.com/support both reach the right team! Sorry for the trouble. We can't look into account specifics here, but our support team can. Open a ticket at netlify.com/support and they'll take a look! 
Launching `a11y` mode in `/design`
Run `/design a11y` to catch:
• Tap targets that are too small
• Focus management problems
• Broken keyboard navigation
• ARIA and accessible name gaps
• Missing or incorrect form labels and errors
Run `cmd update`, then try `/design a11y` in your project 😍
At 12pm ET, Chris Watts from OpenRouter will walk through the hundreds of models now in AI Gateway and how to actually pick one, and Jack Herrington will build a TanStack AI app that swaps models without a rewrite.
There's still time to register! luma.com/c37azk4s GPT 5.6 Sol is now in Command Code GOAT 🐐
Best coding plan with best GPT model. $70 credits.
$10/mo GOAT plan gets you:
GPT Sol has $70 credits
~105M tokens ~2.1K reqs
We've also achieved 99.43% cache hit ratio.
Available on GOAT + above.
Limited time deal.
🐐 
MAN that is a good hero image btao.org/posts/2026-0... All three runs are live. Judge the splurge yourself:
www.netlify.com/blog/one-pro... The expensive run might be the best site of the whole test. It is hard to argue it is four times better than the other two.
The strangest result in our 11-model test came from Claude Opus 5. Three runs of the same prompt cost 253, 249, and 1,055 credits.