Aathil

Which free NVIDIA NIM model is actually fast right now? I built NIM Stats - a live dashboard that monitors NVIDIA's free NIM chat endpoints 24/7 and ranks them by real-world performance. 50+ models · real-time · free 🔗 nimstats.aathil.com

So these labs have more intelligence power internally? So the releases are just slow pushes hmm...

@BarathAnandan7 :)

@AathilOfficial Do not ask me questions which I cannot answer 😭

I’ll give you one piece of advise Grab yourself a 4TB SSD and download it as soon as the weights are out.. IYKYK.. you’re welcome.

@aryanlabde Good question that’s what we call as distribution

claude wrote your code. vercel deployed it. supabase stored the data. now how do you get the users?

@opencode Good, make it cheaper though for now it is consuming a lot

there's a lot of excitement around Kimi K3 so we are making it available immediately to OpenCode Go users we have not yet been able to negotiate a discount yet so it will currently consume your limits at a higher rate hopefully will figure this out soon

Kimi beat Fable? What’s happening? Yesterday someone was saying US Open Source was leading I guess in AI it is always a rapid change huh.?

@roerohan @OpenAI @opencode Use codex for OpenAI models ig.?

If you're using @OpenAI gpt-5.6 Sol with @opencode (or any agent harness that lets the model spawn subagents), you might want to read this. Yesterday I pointed Sol at a debugging task. It spawned an "explore" subagent, so I left it running and walked away. 45 minutes later I came back and it had created 10,409 child sessions. - 10,207 of those were explore agents - nesting went 30+ levels deep - at its peak it was spawning ~350 subagents per minute Total bill: $1,200+ There's not really an easy way in @opencode to notice that it's recursively spawning thousands of subagents while it's running, so I had no idea this was happening. The root sessions barely cost anything (around $18 total). The real damage came from the children spawning more children. Each explore agent kept spawning more explore agents. Usually 2-3 at a time. With 100–140 parent agents doing that simultaneously, the tree just exploded. About 6.7k sessions ended up costing $0 because they spawned and exited almost immediately. The other ~3.7k each racked up a small charge that quietly snowballed into four figures. Until this gets fixed, I'd recommend updating your agent config to disable or at least heavily restrict automatic subagent spawning. This can get out of hand really fast.

@rauchg Cool

Some really cool usecases of Web Analytics API: ▪️ Ask your agent to correlate visitors, custom events (“purchase”, “checkout”), with the evolution of your deployments and performance ▪️ Build custom frontends, and plot this data alongside e.g.: Stripe’s and Resend’s

@jun_song So china is front? And the only companies helping America in this race is OpenAI and Anthropic?

Now Open weight AI models are only 2 months behind. And Gemini, Grok, Meta are at least 6 months behind.

@neerajjj6785 India has? Japan has? Australia has?

USA has Claude USA has Grok USA has Gemini USA has Llama USA has Copilot USA has ChatGPT China has DeepSeek China has Qwen China has Ernie China has GLM China has Kimi China has MiniMax Europe has?

@ItsAlexhere0 Haven’t used codex 20 bucks but Claude 20 bucks is good

one question for vibe coders: which plan is actually worth the money? . Claude $20 . Claude $100 . Codex $20 . Codex $200 . something else honest answers only

@Bhavani_00007 Don’t be so fast in conclusion you never know what OpenAI has in their pockets.

here is how Anthropic wins again : > They're working on Opus 5, and leaks point to Fable 5 tier capability. > OpenAI is offering free resets and temporarily removing limits to keep users from leaving, but it's not a long-term solution. > Once Opus 5 ships, everything shifts. Even if it's not quite Fable 5 level, it still eats GPT-5.6 Sol alive. > And if OpenAI eventually ships GPT-6, Anthropic will already have Fable's next version ready to answer back. > Gemini 3.5 Pro is coming with its 2M context window, but people still aren't sure if the model itself is actually good enough to matter. the winner is already obvious 🥱

@CodeWithAmann Idk, the comment section but I have had good experience with it

I am claude user, scare me with one word.

@svpino Nice question to ponder. Maybe it isn’t token effective after all? Even for the company which created it

Why is Anthropic hiring Software Engineers? Why aren't they using agents with loops to do all of this?

How did my iPhone know today to turn on the DND Schedule by itself after few months of holidays hmm…

@OpenRouter YO, looks awesome but vibe coded but great anyways

OpenRouter has a new look. As we’ve grown into the infrastructure powering the intelligence era, our brand needed to evolve too. Built on clarity, precision, and purpose. For developers. For enterprises. For everyone turning intelligence into impact.

@HermesAgentTips Mimo v2.5, it is made by xiaomi and is good in agentic capabilities (general) but for coding i tend to delegate it to cc or glm via pi or opencode

be honest what’s your favorite model to run on Hermes agent? - Fable 5 - GPT-5.6 - DeepSeek V4 Flash - GLM-5.2

@HuggingModels yeah standard computer with 25gb ram and 370gb storage. hmm., i think you guys need to redefine your scope of a standard computer

Big news for AI on a budget. GLM 5.2 Colibri int4 is a Mixture of Experts model that runs entirely on your CPU. No GPU needed. It's fast, efficient, and opens up advanced language capabilities to anyone with a standard computer. This is a game changer for offline AI.

@moonpay you have 5700 people liking your post, who are you going to send it huh?

sending $500 of Bitcoin to one person who likes this post

@Im_IrushiK fable

Guess the Claude model