Gabe retweeted
Come say hi and build something incredible with @MeshyAI and @UnrealEngine.
Use our new meshy-unreal5.9 skill to guide your coding agent.
@OpenAIDevs , @ClaudeDevs, @SpaceXSI, @googledevs
Meet the 2026 Epic AI MegaJam partners! 🤝
Claude, Google, OpenAI, and SpaceXSI are joining AMD, Diversion, fal, Hyper3D, Intel, Meshy, NetApp, NVIDIA, PBRgen, Reallusion, Rokoko, Tripo, and WeLoveIndies to add tools, credits, and prizes to this year’s jam.
Registered jammers: keep an eye on your inbox. Perk codes and claim links go out today.
See every perk and prize: itch.io/jam/2026-epic-ai-meg…
Excited to support the @opengameio game jam! 🎮
Got a wild game idea? We'd love to see you bring it to life with Meshy. Can't wait to play what you build!
A new game jam is coming from @opengameio...
- $100K cash prizes
- Sponsored by @ElevenLabs, @fal, and @MeshyAI
- 1 month (Nov 18-Dec 18) to show off your creativity
- Build a game using any tool / game engine that exports to the web
Drop your email: opengame.io
i think to succeed as a dev right now:
you must EXPAND your solution space to see what’s possible,
and when working with ai, NARROW its solution space so it hits the target.
in order to narrow the solution space, you need to know what you want to accomplish.
1
90
🕹️ Meshy N1 Texture is live on Meshy web app & API!
What's new:
- Far fewer seams and projection artifacts
- 3× faster than Meshy 7 texturing, in about 20 seconds
- Clean base color at 2K, great for game dev
Try the new texture mode built for stylized game assets!
you still need to hand-hold agents, even the smart ones, for the simplest things.
opus 5.5 was stuck for hours on a simple ui task that i solved for it in ~60 seconds.
on the bright side situations like that make for a great benchmark test for future models.
33
Gabe retweeted
MCP for Blender is an open source success story. Huge credit to @sidahuj for his continued commitment to it.
The Meshy integration opens up even more possibilities. This video brings it all together: MCP for Blender + @MeshyAI + @Remotion + @cap .
recently started working in an entirely new area of our codebase. spread pretty thin so i had 6.1-sol explain high level how the current system works, then i explained how i would go about solving a big ticket. it seemed to understand, so i let it cook.
i even recently built a slick new testing framework that lets it provision nearly any kind of user account setup and run ui tests very fast using the Decisions API. everything passed, we could check off every requirement in the ticket. all seemed well.
had opus 5.5 review it and it just blew everything apart. security issues, poor schema decisions, bad design. and it explained how it would fix everything and i agree with pretty much everything it's saying. opus seems to "get it" more. but i'm not sure how much better it would have done compared to 6.1 sol from the start.
i just keep getting reminders that llms can build stuff that is "correct" but still somehow so wrong. i so badly want to ascend to higher level thinking regarding the product and company, but am constantly reminded that i still need to be in the weeds to understand things and make sure what's being built will actually work and improves the foundation as a whole.
i'm also increasingly of the mindset that consistent ownership of an area of code gives people the most leverage over llms. you're not constantly relearning the landscape and your understanding of the area sharpens every day so you can more quickly assess an llm's work instead of flying on blind faith and vibes.
if you touch new areas of code without understanding things well enough and let your agent thrash around in there, you're likely only going to cause pain. if not now, then later.
spent part of the day yesterday amongst the giants at muir woods.
highly recommend it. especially if you haven’t touched grass in a while.
36
Gabe retweeted
Meshy is offering Hermes users a month of Pro, which includes API access and 1,000 credits, for $1.
After you subscribe and install the plugin, create a key at meshy.ai/developers/keys and paste it into Settings > Plugins > Meshy in Hermes Desktop.
meshy.ai/pricing?utm_medium=…
Hermes Agent now has an official plugin for @MeshyAI.
Ask Hermes to make a 70m poly treasure chest for Godot, or to turn a photo of a mug into a model you can print, and it does the job through Meshy's MCP server.
In Hermes Desktop the models render in chat, rigged characters animate inline, and a gallery stores every model and prompt. On messaging apps, Hermes sends a short preview video so you can create on the go.
hermes-agent.nousresearch.co…
using hermes desktop with the @MeshyAI plugin to build a portal-like game!
incredibly easy and fun. will keep sharing updates as it progresses and put it online for people to play!
Gabe retweeted
3D generation is getting easier and easier to use inside of your favorite tools. Now you can add your Meshy API and start generating assets using the custom UI built for Hermes Agent by 🐐 @Teknium
Gabe retweeted
Hermes Agent now has an official plugin for @MeshyAI.
Ask Hermes to make a 70m poly treasure chest for Godot, or to turn a photo of a mug into a model you can print, and it does the job through Meshy's MCP server.
In Hermes Desktop the models render in chat, rigged characters animate inline, and a gallery stores every model and prompt. On messaging apps, Hermes sends a short preview video so you can create on the go.
hermes-agent.nousresearch.co…
Added the ability to select the model file type to download from Meshy in the in-line chat to the Meshy plugin. Almost there!
Meshy plugin for Hermes Agent going well, full MCP & Skill, plus desktop app integration that lets it display the rendered models in chat, show credits on the status bar, and gives a gallery of all the models you've made
Hopefully releasing tomorrow :)
fuck me. went to micro center 2 weekends ago and was gonna buy a couple of sparks for working on at home.
they had 67 in stock and guy at the store said they don’t seem to sell out as fast as gpus and 67 was a lot to have in stock. so i passed on buying that day.
2 days later, they were sold out. now they’re retailing for $7k. what an idiot i am.
really been enjoying deepseek v4.1 flash. it's certainly not fable level intelligence, or even sol.
but from the right inference provider, it's blazing fast, and actually much more pleasant to talk to.
it seems far more focused on tasks, doesn't write in a way that makes my eyes glaze over, and it almost always provides the exact info i need without me needing to ask for it.
this is mostly based on vibes. i've been working with it for over a week now and it's been my main model since ~last Wednesday while keeping sol as a consultation model for code review or trickier investigations as a sanity check. and i don't find myself getting frustrated at all with it the way i would with claude and, to a lesser extent, gpt models.
maybe i'm behind, but i think we're definitely at a point where a model like this (or maybe even dsv4 flash) is likely "good enough" intelligence for forever.
if model progress just stopped, or the sota models all disappeared tomorrow or i had to stop using them, i'd be perfectly capable and (i think) happy using this model indefinitely.
2
201
scrolling ebay for fun and looking at sold item prices for gpus.
there are way more people getting scammed than i would have guessed.
people buying rtx pro 6000 blackwell gpus for $3k - 4k from accounts created the same month, with zero reviews, and a clearly randomized bot name.
i wonder if people just have dumb bots scouring ebay to find deals and are getting scammed. or if there are just that many desperate or unsuspecting victims.
94
my wife has been in an interior decorating mood lately and using chatgpt to try and re-imagine the space and find stuff. it's been good but a bit of a pain.
she said she wanted to try and build something with AI.
i got her setup with @opencode and using deepseek v4.1 flash through openrouter using together ai as the inference provider for a fast ~100 tok/s setup.
i helped a little bit getting her familiar with the system and how to chat + work with it since she's not an engineer. maybe helped outline how i'd tackle one or two tricky little problems.
then i left her on her own to keep playing with it. came back a few hours later and she had a fully working local proof of concept app. no user creation/management, nothing online. but it's a super clean and polished flow and solves the core pain she had using chatgpt. and it's not the classic chat ui. it's more interactive and creative than that.
it took one afternoon to grasp everything and make something really cool. granted, she's smart, she's got great taste, and has worked in QA before so she's good at thinking about edge cases and testing stuff to provide the agent with feedback. but still, she had never used a coding agent, let alone a tui app.
it also only cost ~$5 total. maybe like $1.50 in coding costs with deepseek, and ~$3.50 in api calls for testing things out.
pretty fucking cool.
reminder - if you let your agent decide things for you, you're leaving a lot on the table and only hurting yourself.
working with gpt sol this morning and asked it to build something for me. it read the code and pushed back saying it wasn't possible, then listed out all the reasons why.
they were nonsense. i countered each point and it changed its assessment. but just to be safe, i asked it to try building the thing exactly like i asked and write tests to prove whether or not it works. turns out it works just fine and was totally possible.
this would have been a missed opportunity or resulted in changes that weren't even necessary if i wasn't already familiar with the code and didn't push back and ask questions.
62
















