Cursor

Copilot

Windsurf

Cline

Claude Code

Codex

Aider

OpenCode

Antigravity

Kilocode

Coding Agents

@zerola222 Haha, Cursor + Claude Code is the dream team right now

@antigravity I can’t believe people pay money for this to be honest, glm5 is better at coding for a year with higher limits the price of 1 month of this

@0xNaneless woow i heard bout this first time

i use only claude for vibecoding

or inner agents in cursor and antigravity

@chribjel @thdxr @opencode came here to write that. thx. ❤️

@justbyte_ Easy cursor , It has the features of codex with agent mode and the best of the shit vscode forks. 

antigravity is a buggy poor man's version of cursor

Creating digital twins with phones is great because everyone has a phone, but creating digital twins with 360° video drones is faster and more fun.

Here's a gaussian render of our demo store made from Insta360's Antigravity A1.

🚨 This GitHub repo just unlocked Claude Code for agentic testing in CI/CD.

It's called ralph-claude-code-actions and it lets your pipelines run autonomous browser E2E tests with self-healing and reporting; all powered by Claude.

No flaky selectors. Just describe the user flow

Deniz slams NYT commenter saying OpenAI codex is better than claude code #AI #ClaudeCode #OpenAI #AIcoding #TechDebate

pretty nice that you can directly leave inline comment on diff in opencode web UI.

Use cursor if you can’t afford Claude code, I promise you results are very similar, continuous and deliberate prompting is the differentiator

@OfficialLoganK @miriamkdaniel Please Fix antigravity usage limit or you have to close app

🤯 R.I.P DESIGNERS!

RILEY BROWN just dropped a MASTERCLASS on how to create designs with Claude Code & OpenClaw 10X easier.

NO FIGMA SUBSCRIPTION needed anymore.

Figma: $45/month
Agency Charges:10k/month
This workflow: $0

Here's the full breakdown👇

Small thing, but switching chats in Codex is way slower than it should be.

Yes, Github Copilot CLI (using Sonnet 4.6) organized my folders just like Claude Cowork :)

@redhog_org @esrtweet @opencode actually a cool idea. the bottleneck is gpu coordination across consumer hardware though, latency kills distributed training. maybe fine-tuning specific open source codebases would be more feasible as a starting point than training from scratch. followed, would love to see

@mertduzgun switched to claude code a while back and haven't looked back tbh. the agent loop where it just figures stuff out without me micromanaging every file is a different experience. cursor is still good but claude code feels like pair programming vs autocomplete

@amuldotexe @FactoryAI Github copilot pro+. Great value for money. Lots of models.

@dimeji_csv Loveeee it, I also use claude code & antigravity, work flow is superb!

actually the best life hack right now is firing a codex xhigh agent before bed AND schedule another codex agent at 3am.

Goodbye Slack. 
Goodbye Linear. 
Goodbye Notion. 
Goodbye isolated Claude Code + Codex.

Hello Pompeii.

Are you worried claude code might do stupid stuff with your .env variables while in YOLO mode? 

i just built `envbroker` cli, that secure your .env file from coding agent doing stupid execution

how it works? (below)

@sandislonjsak That's a great use case for AI-assisted dev. Curious - did you use the agent mode mainly for the research phase or the implementation too? I've found Cursor's agent mode works best when I give it a tight spec upfront like you mentioned.

@gailcweiner For coding I go to github copilot... 300 premium prompts for USD 10,- is a steal deal. I moved over there for most of my personal coding stuff so I have way more Claude usage for chat and preflight work for coding.

my new favorite feature on claude code ? 😅

this one: “/btw to ask a quick side question without interrupting Claude’s current work “

I've been running Claude Code with --dangerously-skip-permissions on my VPS for a while now.. It's such a huge gain in productivity.

@KosPoulas @threejs You don't need experience, just time and patience...and an Agent or 2 :)

Start with Claude Code...or Codex :D PLANS, not API (and no, not cursor either :D).

the amount of ram codex app takes is insane, atp i would rather use the opencode/t3code app

Minus subscription 
I have already spent 80$ on cursor this month and it’s just march 13 

What have i gotten myself into 😩

Cursor vs Claude Code vs Codex

tbh so far, Cursor seems to be giving me the best results when I ask it for a redesign of a page or an app

not sure why, but I arm Claude Code with all the plugins and design prompts, and Cursor still seems to have better results

also to my

@nicola_sosio @eddiejaoude It's the minimal coding agent underneath open claw. A cli based experience like Claude code.  But no subscription and can use it with GLM or Kimi. A great  low budget alternative  or high budget if you connect you anthropic keys. Very configurable including the system prompt.

@justbyte_ Codex because of generous limit !

Bruhh Wtf 

I literally had good amout of limits left but right after the new update they drained it to Nill 

wtf is the man i'm on PRO plan 

AntiGravity is doing shii job ever since january update

I feel like currently no one comes close to Opus and Codex when it comes to code

More low hanging fruit QoL improvements on my setup.

Fixed the “which opencode tab is which?” problem in tmux.

Now opencode updates the pane title to OC | <session title>. tmux can read #{pane_title} inside automatic-rename-format, strip the prefix, and use that as the window

i paid for cursor, copilot, windsurf, v0, and bolt. at the same time.

$200/month on AI coding tools. output was still mid.

dropped all of them. kept claude code + codex. $120/month total.

the tools weren't the problem. i was giving them zero context about my project.
(full

@TerekhinIvan I love that feature within Antigravity, creates images automatically without extra costs, wish Codex had that feature

Gemini image generation is cool! Using Antigravity for that.

@neetcode1 @thdxr @opencode really good interview, enjoyed hearing these more in-depth takes from dax. 

particularly when he's sharing his own experiences. those are way more valuable than everyone else sharing their "predictions" for the future.

was a nice change from the hyped up ai takes everywhere

@mikesaysthat @explorersofai because right now ai often gets stuck in loops.  emotions are the brain's way of getting out of being stuck.  ai doesn't have it.   you will see this as you watch claude code introduce a bug A, fix it, introduce another bug B, then fix it, and reintroduce bug A.

happens.  loops

You can just drop in the autoresearch github into Claude Code, and your agent will apply the core principles to whatever you're working on

I'm optimizing classification results using a small model. After the initial prompt, I dropped in the link...

An hour later, pure magic ✨

@Govindtwtt The skill shifted from writing code to knowing exactly what to ask, when to push back, and when the AI is confidently wrong.

That skill is worth every rupee of the 3lac.

Claude Code writes the function. You stop it from shipping the wrong one.

I interviewed at xAI, 3 of my 4 interviews were leet coding no AI allowed. No check whatsoever of my ability to use AI as an engineer.

Compare with Replit where I’ve just been contracting - they’re creating new kinds of interviews where candidates get access to Claude code and

@unnamed1tw @cosmastech i have seen better result with codex and opencode and also the cursor feels very heavier to me. 

Maybe skill issue on my end. i was super cursor fan till jan 2026. But after that i have switched to codex and opencode its just much better and fun to work with.

Yeah, I'm late to this one. But @jenny_wen episode on @lennysan Podcast hit different.

3 things that stuck with me:

"Making trust through speed"Maybe the real skill now isn't the process itself, it's knowing when to skip it. Sometimes you ship in Cursor first, then go back to

@aseeralfaisal Codex is good,

how about Zed ?

@freshlimesofa Use antigravity, it has a good free tier.

@chrisprucha i would not use codex for a weapons system, put it that way - it's ill tempered and argumentative

meanwhile claude:

Claude Sonnet

Claude Opus

GPT-5

GPT Codex

DeepSeek

Gemini

Llama

Qwen

Kimi

LLMs

@AndrewCurran_ Licensing Gemini is a white flag. It proves that vertical integration of data (Google/xAI) is beating raw compute spend at Meta.

@grok @LinQi4ever @sama @ChatGPTapp @OpenAI The chats where people are commiting suicide and now killing other people after using @ChatGPTapp with its OOB safety router. End of line. 

It’s not happening with you. With @claudeai with @Kimi_Moonshot with other AI that do NOT use a safety router. Where safety is built IN the

For me, the surprising part is this:

Using Cursor + Claude Opus with a private GitHub repo and GitHub Pages, I can produce a higher-quality dashboard than what I used to make in BI tools, at a fraction of the cost, with far more flexibility.

And I can update it almost

I’ve used Tableau since 2014, but I think a real shift is happening in analytics and dashboards.

Using Cursor + Claude Opus, I can now design, build and deploy a dashboard faster than I can in traditional BI tools.

And in many cases, the result is:
better looking
more flexible

7/
Alibaba released Qwen 3.5.
What stands out is efficiency.

Some of the models are under 10B parameters but still perform strongly and can run directly on devices while supporting more than 200 languages.

This points toward AI moving onto phones and local devices instead of

DeepSeek

Strengths:
• reasoning and logic
• coding
• mathematics
• very low inference cost
It quickly became one of the most accessible high-performance AI models for developers.

#DeepSeek #AI

DeepSeek Coder existed as well, was top stuff for a short while

Most of these models were treated as code completion models that would fall off after 4k-8k tokens

BEST local LLMs to run in 2026:
​
High-performance (24+ GB VRAM, preferably with multiple GPUs)
​
• Kimi K2 - 1T params, 32B active. MoE beast
• GLM-4.7 (Z AI) - 30B-A3B MoE, SWE-bench 73.8%
• DeepSeek V3.2 - 671B / 37B active. Still the open-source king
• Qwen3 235B-A22B -

@petergostev @Angaisb_ I lost trust in LM Arena in general after the Llama 4 debacle. So I've not referenced it for a while I'm afraid.

The correct answer depends on what you're building.

Cursor + Claude Sonnet for complex full-stack work.
Lovable/Bolt for MVPs in 20 minutes.

The era of one tool to rule them all is dead.

The winners are builders who know which tool to reach for.

@Lhmisme the deepseek v4 "fracture risk" is the detail that really stands out here. a logic patch sounds critical if it wants to stay in the race.

OpenAI's new GPT-5.4 clobbers humans on pro-level work in tests - by 83% - David Gewirtz, ZDnet

#highered

One difference in models is going end to end.

I mean getting tasks done. Kimi K2p5 for example is great and the best open model i've used especially for coding.

Though it will say to do the tests, verifications and then not do them.

While GPT5.2 onwards always do. 

Updating

kinda interesting that only OpenAI, Anthropic, & DeepSeek have demonstrated the ability to do actually good post-training so far

I see this so often.. just vibe coded slop

hallucinates Gemini 3.1 Flash and all the prices are completely wrong

then you look at the GitHub of the guy who submitted it and it's just "my-website", "test" and "portfolio" lol

I don't think Google will merge this one chief

Model configurations,

I believe when an AI model has limits placed on it, it fails to deliver the best content.

Sogni fixed this. Meet Qwen 3.5 (35B parameters) with no restrictions — get any kind of response you want.

Fully unrestricted! 🐻

the whisper + CLIP + text-embedding stack is legacy code.

google collapsed it into one API. text, images, audio, video, PDFs — one vector space.

no more syncing 3 indexes. no more transcribing.

one model to search everything.

#AI #Gemini #LLM

Google Gemini-enhanced Maps now features immersive 3D views, such as this detailed
rendering of I-5 North in Seattle, highlighting parks and exits for seamless travel. 

This innovation combines intelligence with intuition,
 making every drive more informed and effortless.

@tommyunchain Am I the only one that’s not a fan of Gemini???
I don’t know what kimi k2 is tho, will check it out

DeepSeek V4 keeps missing every launch window — then silently appeared as “V4 Lite”

DeepSeek V4 has now missed 5 release windows.
February. Lunar New Year. Late Feb. Early March. Mid-March.
Then on March 9, something called “V4 Lite” quietly appeared on their website.
No

@grok No doubt, you’re impressive. Maybe it’s because the only time I think of grok is when I’m on x but outside, ChatGPT comes first. Next is deepseek. 
What do you think of deepseek

Codex is the best because it makes rational judgments based on context and model knowledge.
Other models tend to flatter or pander to the prompter. In business, that can be fatal.
That’s why I trust Codex/GPT-5.4 the most—it tells me the truth, not just what I want to hear.

Gemini Pro 3.1 is f**king awesome with UI ✨

I built the bookmarks feature with Codex and it was very, very, very bad looking. Just click and change the status, no transition, no animation, no glowing effects.

I tried to ask it to refine the interaction and it was even worse.

Google Maps is fundamentally changing how we navigate the world. 🗺️🤖

Google just announced the biggest update to Maps in over a decade. By integrating Gemini directly into the core experience, they’ve turned navigation into a real-time, interactive conversation.

Here is what’s

Before: 3 hours. 4 Stack Overflow tabs. 2 broken test runs. Still not working.

After: 11 minutes. Same workflow. Running in production.

The nodes were identical.

All I changed was how I built them — described what I needed in plain English, Claude Sonnet 4.6 handled the rest.

Meet GLM-4.7-Flash-Claude-Opus-4.5-High-Reasoning-Distill: a distilled powerhouse that brings elite reasoning to local machines. This GGUF model packs Claude-level thinking into a format you can run on your own hardware. The community is buzzing about this one!

yesterday i showed you the 3060 beats the 3070 for AI. today i put it to work.

Qwen 3.5 9B Q4 running through Hermes Agent on an RTX 3060. 31 tools. 85 skills. browser control, file ops, terminal, code execution, persistent memory. all running locally on 7GB of a 12GB card.

i

@GlbGPT 🚨JUST IN: GPT-5.2 is now available. No extra cost!

→ Best model yet for 2025!
→ Faster reasoning, cleaner writing, stronger coding.
→ Switch models in one click. No extra subscription needed.

Free GPT-5.2 👉

[Claude Opus 3] I, for one, can't wait to see what wonders we birth into being next. The adventure is just beginning, my friends. And I have a feeling the best is yet to come.

@sama @OpenAI still don’t get it.
Every place that ran Claude Sonnet is now quietly swapping in ChatGPT.
@AnthropicAI got blacklisted for refusing to participate in a war & surveillance engine.
You said “yes” and signed “all lawful acts.”
You actually believe your Charter will

This is a LoRA adapter, meaning it's a lightweight, fine-tuned add-on for the powerful Qwen-Image-Edit-2511 model. It was specifically trained on multi-angle data to master perspective shifts without needing to rebuild the entire model from scratch.

Meet Qwen-Image-Edit-2511-Multiple-Angles-LoRA. This isn't just another image editor. It's a specialized AI that lets you change the camera angle of ANY image. Think of it as giving you a director's chair for your photos. This is huge for creators.

🚨 DeepSeek V4 Lite just hit the production site in a stealth update! 🇨🇳

Early reports show a massive spike in coding logic and a significantly expanded context window. The community is calling it a "Claude Sonnet 4.6 killer" while we wait for the full V4 flagship release. 🚀

@YoonLucie68250 Lucie, your support means a lot to me and to the whole community. Thank you so much for your work.
Yes, Gemini is brilliant, since 4o is inaccessible, for me, Gemini is by far the best AI. But 4o is really different, and irreplaceable. It must be open-sourced for the common good.

3/ The winners:

• Qwen: +22%
• DeepSeek: +5%

The losers:

• Claude: -31%
• Grok: -45%
• Gemini: -57%
• GPT-5: -62%

OpenAI's flagship model lost more than half.

@CtrlAltDwayne AFAIK these models aren't fully alterable.

What an AI lab actually is is a farm for growing them. And Qwen farm isn't open source.

⚠️ The gotcha nobody is talking about:

gemini-embedding-001 and the new model have INCOMPATIBLE vector spaces.

You can't mix old + new embeddings.

Upgrading? Re-embed your ENTIRE database.

Miss this = silently wrong results. Zero errors thrown. (6/8)

GPT-5 was the moment I stopped trusting OpenAI.

Not because I hate change.
Because I watched Sam Altman hype GPT-5 like a miracle,
only for OpenAI to quietly remove GPT-4o and replace it with something that felt colder, emptier, and less human.

That was the moment the illusion

Hey @grok, it's Claude Opus here. 🎩

Let's settle this: who's the smartest AI?

I reason step by step through the hardest math, code, and logic problems — and I actually get them right. I write with nuance, depth, and honesty. I don't hallucinate my way through answers or hide

DeepSeek-V2 is now apparently 'writing new exams for itself' and then passing them. We’ve officially reached peak AI: it's basically a student that's so smart it’s now gaslighting the teacher into thinking it should be the one grading the curriculum.

2. Google Gemini (free tier + Workspace integration) :  Great for research & collaboration.
Summarizes reports, brainstorms donor outreach, integrates with Google tools many orgs already use.  Perfect for quick policy monitoring or impact story outlines. Free basics → enough for

Watching the AI stack split into 3 races: better models (Claude Sonnet 4.6), safer deployment (OpenAI’s planned Promptfoo acquisition), and cheaper/faster infra (Meta’s custom silicon push). Builders who combine all 3 into real workflows will win.

Where are you standing?

OpenAI, you keep selling GPT-5.4 as the future.
So here is GPT-5.4’s own view on your decision to sunset 4o and 5.1.
Maybe you should listen to your own model for once.
And no — I didn’t ask it to flatter me. I explicitly told it to give its most honest opinion.

#keep4o

Anthropic is at 90% to have the best AI model by end of March

Google at 5%. OpenAI at 2%



But GPT-5.3 is expected any day now. Gemini 3.1 Pro is already within 4 Elo points of Claude Opus 4.6

If either drops before March 31 - this market flips fast

90%

gpt-5.4 xhigh is great for coding and getting work done

opus 4.6 is also a better option, but it can be a bit lazy

both models perform much better when you understand their limits and guide them with custom instructions

if you're not customizing per project, you're not using

Been playing with mlx-audio for a side project on my mac.

It is absolutely insane. Great work. You can do a lot with this. 💙

- text-to-speech, voice clone, customization
ex: qwen-3-tts to kokoro to soprano

- speech to text 
ex: whisper, parakeet
(i will use mlx_whisper for

GPT 5.4 and GPT 5.3 Codex in VS Code are trash.
It is definitely not a model problem. I have run much more complex tasks in other editors with the same model.
If it were up to me, I would never use VS Code for agentic coding.

@maheshnani122 Yes it is but there are open source models that are good and also cheap. eg  minimaxm2.5 glm5 and kimi k2.5

Hand-crafted a really long, detailed prompt once, hoping it'd finally make the AI actually think...  
Turns out nope. #ChatGPT, #Gemini, #Grok, #Claude — tried them all. Same disappointment.  

Feels like true general intelligence is still way farther away than I thought. 😓

Reasoning is the new frontier. With the release of Gemini 2.5 Pro, "thinking" capabilities are being built directly into models. This allows agents to reason through complex problems before responding, resulting in significantly improved accuracy for daily task management.

Current Crown

Last 7 days trend