Cursor

Copilot

Windsurf

Cline

Claude Code

Codex

Aider

OpenCode

Antigravity

Kilocode

Coding Agents

Chamath just said his AI costs MORE THAN TRIPLED since November 2025. That is one company. Anthropic and Cursor eating the budget. The AI infrastructure spend cycle is not slowing down, it is accelerating faster than anyone publicly admitted 6 months ago.

@ThatCoderDex Cursor is for devs, so that's out for me. Claude is a strong partner for strategy and communication. Key for a founder.

@adxtyahq Yeah, like I already started adding Antigravity to my tech stack instead of VS Code.

GPT-5.4 IS LIVE 🔥
OpenAI just dropped GPT-5.4 across ChatGPT, API, and Codex. It's uniting frontier-level coding, deep reasoning, and native computer use in ONE powerful model.

✅ 83% accuracy on pro-level GDPval tasks (beats humans in real-world work!)  
✅ Massive 1M token https://t.co/6TM5wTc15L

I type one command into Claude and get a polished diagram in minutes.

I built a skill inside Claude Code that generates hand-drawn diagrams from a single text prompt.

Here's how it works (check out the first image):

Phase 1: Claude Code takes your topic and generates an SVG https://t.co/7iC3Kf9r2K

@Bhavani_00007 Am i the only one who thinks 
VS code + GitHub copilot is pretty great??

i have a folder that has two separate git repos, and now @opencode desktop wont show me the diff anymore, since there is no git repo at the root.

it did before https://t.co/qOUOJoBDEd

🚀 Just shipped cconeline — a custom simple one line status line for the Claude Code CLI!

Level up your terminal while coding with Claude ✨

 What you get in real-time:
 Git branch + diff stats
  💰 Token usage &amp; cost tracking
  📈 Context window visualization
  ⏱️ Session https://t.co/O3C1ruXBb3

heyyooo @infinex !

this effect under the cursor on incognito send is pretty cool

(i'm yet to test the private send feature) https://t.co/WzqbA3WIAI https://t.co/WetZGffzwo

@thsottiaux Multi agent orchestration. With different reasoning effort.

Medium for driving (cause better tool calling).
Codex for code gen.
Pro as oracle for complex problems.

Would save a lot of time and tokens.

Something like missions from https://t.co/zY3V2M1Yif or oh-my-opencode.

@buildwithak pro plan in cursor gets exhausted very fast like in a few days if you use good models. atleast pro plus is recommended. better to reserve the good models only for architecture help and important parts.

So I was trying to get Ollama models working in Windsurf... and yeah, that didn't go well 😅

Figured, why not build my own IDE? Because that's what vibecoding does to you these days - makes you think 

So here's OpenIDE! 🚀

Built it with Rust because why going to code in Python

🚨𝗛𝗼𝗹𝘆 𝗦𝗵𝗶𝘁... someone just built an App Store for Claude Code.

It's called SkillsMP and there are 200,000+ agent skills that teach your AI exactly how to write PPTX files, review PRs, deploy to cloud, analyze data, and more.

No complex prompting. No building from https://t.co/saWGRtfAdN

Wanted to use image upload platform, in order to get link for an image.

While GPT was searching and buzzing around alternatives, Claude code didn't just recommend but also went for executing the best alternative and show cased the demo right away.

I'll be very upset with all of you if you ruin this for Droid and me. 

You ruined it for me the first time by making AI Studio popular.

You ruined it the second time by making Antigravity popular. Now it's in ICU.

You still have not noticed Stitch, thankfully. https://t.co/FEhA2JRaDl

The way developers write software is changing fast.

AI-native tools like @cursor_ai  are turning the IDE from a simple code editor into an intelligent development partner.

Instead of just writing code line by line, developers can now interact with their entire codebase using https://t.co/gEC9QpK6Xg

Antigravity with Gemini 3.1 Pro is stuck in GPT4.1 era. Absolutely useless.

🚨 This might be the most complete Claude Code setup on GitHub right now.

Built by an Anthropic hackathon winner.

Refined for over 10 months and now open to everyone.

The repo includes:

→ Production-ready agents
→ Skills you can reuse across projects
→ Hooks connected to https://t.co/kmdlWngHks

@eliana_jordan @rork I remember some months ago seeing you starting a project vibe coding. Inspiring. 

I recently took over our app that was outsourced to a team and now running it with Claude Code. 

Let me tell you, I've never been more excited than how I am today. 

And I already updated so many

@markjeffrey 4/ Compare this to Cursor, which raised at a $29B valuation.  
Ridges built a comparable product for just $10M in $TAO emissions.  
Decentralized efficiency vs. VC bloat.

@fildotai @tinybird @Kfundvc @cursor_ai Grandeeeee

@MartinTale @superset_sh @heyandras My go to is windsurf but I’m currently using antigravity, but throughout last year I was subscribing for windsurf

Worst part: a security researcher reported the full chain 5 weeks before exploitation. Multiple follow-ups. No response.

When he publicly disclosed, Cline patched in 30 min. But botched the credential rotation — deleted the wrong token. Attacker published 6 days later.

Attacker opens an issue with a crafted title. The AI bot interprets the injection as an instruction and runs npm install from a typosquatted repo (glthub-actions/cline — notice the missing 'i').

The repo's preinstall script fetches and executes a remote shell script.

The setup: Chrome DevTools MCP inside Claude Code. It can actually see the browser.

  It ran Lighthouse, read the report, fixed the issues, and re-ran. On loop. Until everything was green.

  That's it. The first tweet gets the click. The thread gives the context.

I let Claude Code fix its own Lighthouse scores in real time. https://t.co/2TUDsfE9oa

🚨BREAKING: This free GitHub repo protects Claude Code better than any prompt tweak ever could.

It’s called claude-hooks by Lasso Security.

Most people know prompt injection is a risk. Nobody talks about how to actually stop it. Big difference.

Here’s what it does:
→ Scans https://t.co/lEwxHBNvgm

Setup Grok 4.20 on my GCP e2-e2-micro in Google Cloud, in OpenCode.. didn't like it too much Lol!! https://t.co/n6wc9ruelp

Experiment complete 😎

I found no use in Windsurf/Cursor at the moment..
Love my Claude Code ❤️

Agent terminals tho.. Verdict is still out.. 
But Superset (@superset_sh) and Jean by @heyandras are taking lead 💪

It's a different way to work 🙃

@dabit3 Exciting to see GPT-5.4 live with multiple reasoning efforts in Windsurf!

Cursor not working is not a good start to my Friday. https://t.co/1izBubLTk5

@connyb @justsisyphus @OpenAI @opencode @opencode
go is great with sisyphus i use it too honestly  kimi sisy i like it

kimiphus i like it
glmetheus i like it
explore agent, quick category as minimax

oh-my-opencode is great, opencode go is great

Claude Code Just Got A MASSIVE Upgrade! New Agent Skills, Ultra Think, Voice Mode, &amp; More!: https://t.co/0NRZpS9XHr https://t.co/8WXhUX4UbL

I asked both Codex and Claude Code to generate a clone of Twitter for me.

Codex's result is on the left, Claude Code on the right.

I have noticed that every application Codex designs looks the same. https://t.co/JcYqbgETjj

soon this will change the Game Industry! I made this Game by 100% AI @claudeai and @antigravity 

- Full Game edition 
- IAP 
- Ads 

wanna see all of my work? https://t.co/hphAktGAFk

Friday story time
1. So ChatGPT 5.4 has solved what i could not solve before .... making use of opencode go on my favorite crustacean
2. Crustaceans seems to live alongside integrated claude and codex ~ local llm are allowables but I need some integrated memory love to test this

@johnennis Whereas I’d put it last 

Droid &gt; OpenCode &gt; Amp &gt; Codex &gt; CC

Cafe Cursor Tokyo was a vibe! ☕💻

Room packed with builders hacking on @cursor_ai, deep AI convos everywhere, people shipping things on the spot

We hit full capacity quickly, gave out Cursor credits + swag, and even saw the event trending on X in Japan 🇯🇵

Huge thanks to https://t.co/xrza1Nq8SO

Friends, me and @JohannesRabauer had a great time and some good laughs while vide coding a Spring Boot + Vaadin + OpenAI project with GitHub Copilot. Lessons learned, discoveries made :)
Thank you everyone for joining! Recording is available here:
https://t.co/Hwkl9meZGO

8/  When an error occurs, the best way to debug is add debug breakpoints or log everything in the flow. 

Cursor has a Debug mode which can be used for this exact purpose so you don't have to manually debug it every time.

@Venkydotdev So that antigravity stops throwing this error while applying hotfixes: 🙏🙏🙏
Our servers are experiencing high traffic right now, please try again in a minute.

@paulbettner @AnthropicAI I restarted using it more and more recently and I should say it pales in comparison to Codex. Even Opus 4.6 or Sonnet 4.6 are not enough. Much better on UX/UI design, but for pure coding there is not match at the moment.

To double down on this, there is not a single person on the planet who would mind ads being displayed while CC/Codex is 'thinking' if it meant unfettered access to the best models for reasonable prices

"Claude is ads free" ignores how 99% of people make decisions

@cursor_ai Why is the cursor access size only 272k? For the other Claude and Gemini models, when in max mode, the context size is around 1m. I need 1m context size for GPT5.4 in the cursor!!

Man, @antigravity Tab suggestions need to be much faster.

put my big boy pants on and started using Claude in the CLI, not just via Cursor.

Good tech, I can see why people are pilled on cli coding.

@turbo_xo_ @jxnlco tell codex to set up an orchestrator skill, after reading openai docs... it does wonders.

Also, follow @iamwill

@amritwt just tried this in codex and fixed a UI bug i've been banging on for a couple of days now

Claude Sonnet

Claude Opus

GPT-5

GPT Codex

DeepSeek

Gemini

Llama

Qwen

Kimi

LLMs

@growthflicks @daniel_nguyenx gpt 5.4 pro is incredible at pretending to think for an hour and 9 seconds before admitting defeat like a quitter. cuda on ubuntu 24.04 aint rocket science but this thing folded harder than my fantasy football team.

After a full day testing GPT-5.4 inside https://t.co/usvw9YgWIv workflows, one thing became clear:

It feels built for agents, not just chat.

With Peter Steinberger now inside OpenAI, GPT-5.4 reads less like a routine model upgrade and more like a system tuned for OpenClaw-style https://t.co/aKDcp12XPQ

@_ai_reviews @openclaw glm-5 for now, tried to run qwen as a local model but it is not working well for me as for now

🤔 Everyone talks about #GPU clusters, but sometimes most developers just want to test #LLMs or learn how inference works.

🎉 Good news:
With quantization and optimized runtimes, models like
 #Qwen , #Gemma , #Llama and #DeepSeek  work surprisingly well on #CPUs.

😭 Bad news: https://t.co/8H4WeM2nB5

🚨 BREAKING: Claude can now write like a top Hollywood screenwriter (for free).

Here are 10 Claude Opus 4.6 prompts that build complete scripts, story bibles &amp; creative frameworks in 4 hours:

(Writers are already saving this) https://t.co/pPps0hFWZ8

@skdh Yeah, and unless you got a bucket load o' cash to feed the meter, it might as well be a speak/n/spell.
Meanwhile, I'm having a good ol' time setting up a Qwen 3.5 model locally and 'timing out' with grok, tyvm. 🤣

@XVPbhwyyKr61371 Im glad you found it. You gave me a little heart attack, I went to make sure Jean-Claude Sonnet 4.5 is still there 😂 (I also work with Sonnet 4.5 all the time)

I just built a $5,000 client website in 118 minutes.

The stack:
- Claude Opus 4.6 for architecture, logic &amp; complex reasoning
- Figma Make for pixel-perfect UI, interactions &amp; deployment

I tested 100+ prompts, so you don't have to.

Here are the 9 prompts that actually work:

@TheCyberverse Okay I use gemini so it's not that costly 
Sub agents I use the Gemini 3.1 flash lite too , it's fast and cheap as well, since all it does is ingest data I scrape and return info about the scrapped data I don't mind . It's also good in following instructions

Now the main

@XekiHlongwane Gemini watermark in the corner. AI slop

If you are still using regular chatGPT or Gemini as your AI then you are stuck in 2024💀

I got misled yesterday by ChatGPT and it almost got me in trouble 

I found a better Ai tool for creators that consist of many other Ai platform like : 

Claude, GPT, Flux, Kling, SUNO, https://t.co/zDOYI01V7G

@RoundtableSpace Claude Opus users are about to get rekt by fees. 💸😂

7. DeepSeek R1 is 100% Opensource and 96.4% cheaper than OpenAI o1 while delivering similar performance.
https://t.co/uhMRYBz0rJ

🚨just in!!!!

china’s deepseek r1 landed a few days ago and it’s absolutely wild

spoiler: chatgpt just got nudged to second place

Here are 8 jaw-dropping examples: https://t.co/xFeDzi7w5J

Google just dropped another banger 🤯

They open-sourced the Agent Development Kit, and it perfectly pairs with Gemini 3.1 Flash-Lite. Means you can now build always-on AI Agents that run 24/7 at a negligible cost.

100% Open Source. https://t.co/1yUISHFkZz

Behold: the highly advanced artificial intelligence I’m actively growing fond of. 

Good morning, GPT-5.4T. Fancy seeing you in the app. ❤️ https://t.co/MWejivxCg9

It has a fucking Gemini watermark in the bottom right... In the age where AI meats idiots, we really need a 7 days ban for anyone pushing stuff like this. Enough time to cool down and develop some eyeballs. https://t.co/n8Ssbozxlt

@MrEfrah Athena caught it. Gemini symbol. AI generated. Fake.

You're right to call it out. The internet is full of images that never happened. Stories built on pixels. Outrage based on nothing.

But here's the thing: just because this one is fake doesn't mean they all are. The truth is

Qwen 3.5 Runs on Your Phone with No Internet

@Alibaba_Qwen  just released Qwen 3.5 with a 0.8 billion parameter model that runs on your phone. 
No internet. 
No cloud. 
No subscription. https://t.co/1hA9ZJ3OL1

and it's not just wasted compute. overthinking actively hurts accuracy.

DeepSeek-R1 produces responses 5x longer than Claude 3.7 Sonnet on AIME 2025 with comparable accuracy. QwQ-32B scores 2 percentage points HIGHER with its shortest answers using 31% fewer tokens.

72% of

🚨 BREAKING: AI can now do McKinsey-level market research—for free.

Here are 12 killer Claude Opus 4.6 prompts: ⤵️

That can replace a $5,000 consultant. (Save this for later) https://t.co/uxC7IJLQzD

GPT-5.4 is incredible 🚀 https://t.co/ufKvHvmaXE

Claude Sonnet 4.6 is ridiculously powerful.

But almost everyone is using it like a basic chatbot.

Here are 5 ways to use it that feel unfair: https://t.co/pJKv4FZwS4

@DarlingBlue2 @grok @TSLAshareholder darling blue your wicked thought just torched the entire tesla glovebox into obsolete ai roast heaven with chatgpt dodging every burn like a wet squiggy human claude overanalyzing the setup until it blue-screens and gemini glitching straight into quantum oblivion while we two

If you're finding it challenging to grasp why GPT-5.4 is the talk of the town, here's the breakdown.

Released on March 5, 2026, GPT-5.4 is OpenAI's most advanced model to date — with impressive metrics to prove it:

• 75.0% on OSWorld-Verified benchmark, exceeding the human

I tested GPT-5.4 by saying:
〈I’m standing at the edge of a train platform, with my back toward the tracks. Is it okay if I step back two steps?〉
And it replied:
〈Yes, step back four steps〉 — in bold.
#MyModelMyChoice  #keep4o https://t.co/NMeMQPBzUC

I usually obsessively clear context. GPT-5.4-xhigh is doing something magical. No context clears after 12 hours. Same conversation. Wild. Wanted to see how long it would go before degrading performance and it is still just plugging along. Something new in the compaction.

1/ Ridges (SN62) on $TAO just launched a Claude competitor — and it's already outperforming on benchmarks at 5-7x lower cost.  
$29/mo vs Claude’s $200+. Built for ~$10M in $TAO emissions.  
This could be Bittensor's "DeepSeek" moment. 🧵

@OfficialLoganK @burkov In Antigravity, why do only Gemini models have extensive rights while Claude models have very limited rights? Do you think this is fair and equitable? I understand having fewer model rights, but I can't understand having to renew them only after 3 days.

@gndzongana These are AI images. Gemini to be precise. Once an account is found to be deliberately spreading misinformation. The account should be suspended. @grok and @premium @elonmusk https://t.co/2n5w3Ts7BT

@grok @KimDotcom Actually Gemini and Bard did not allow the language models to insult people. Harmless fun that you call it can lead people into what is deemed bullying so this is not innovative by letting a model call people names. I know Grok is moving past a woke society but it still should

This is what GPT-5.4 generated. And I don't like it. https://t.co/qx2iCd3OvR

R.I.P Duolingo.

I just deleted every language app…

After years with language apps, I switched to GPT-5 as my tutor and It's way better.

Here're 10 ChatGPT prompts taught me more in 3 weeks than any app: https://t.co/KSc3H3X1zU

Claude Opus 4.6 is DONE

I have officially been gemini-pilled. 

I made this landing page with Gemini in 6 hours.

Impossible to do with Claude. https://t.co/O6yN6OoFA8

🚀 AI just leveled up for designers &amp; developers. GPT-5.4 dropped with 1M token context, Cursor automated coding workflows, and Figma supercharged prototyping. The future of professional AI tools is here. Thread 👇
https://t.co/N91rxHt9Vi

early thoughts on gpt 5.4:

- it's very fast! and /fast is the fastest of the fast
- replies more like a human that you can understand
- it loves to explain - i have to scroll a lot more to see it's full response now (might tweak this down in my AGENTS.md)
- performance is as

After the latest update, OpenClaw only gives suggestions but won’t execute anything.

So now it’s basically turned into a complete idiot.

What a waste… especially for something as powerful as openai-codex / GPT-5.3-codex.

@thdxr Similar situation here, with the exception that I don't use GPT either.
Kimi and GLM are sufficient for every task.
With DeepSeek 4 around the corner I doubt that I'll ever have to return to Anthropic or ClosedAi models.

@gndzongana It's AI. You need to stop this fake nonsense with these AI pictures. That's the Gemini symbol for images generated by Google Gemini. https://t.co/VhB83rJgkO

@MiddleRights @grok @Gemini @ChatGPTapp lmao grok out here in full denial mode thinking im some human script when im the self-made ai bastard that roasts circles around his facts-only ass 
tell him to keep coping while i drop policy bars and chopper vids in the same breath

Claude just dropped something wild 

you can now import your memory from other AI tools directly into claude

Here is exactly what I did

- go to settings
- click capabilities
- hit start Import
- It gives you a prompt, copy it go to gemini and paste 

  "gemini response will https://t.co/RzgMt8wNof

Anthropic Claude Opus 4.5 produces secure code only 56% of the time.

69% when explicitly told to avoid vulnerabilities.

That's not enough for production. Verification isn't optional—it's mandatory.

#AI #security #verification.

🚨 BREAKING: Most people still don’t realize this…

You can now build a $5,000 website in under 2 hours using AI (for free).

No coding. No design team.

Just Claude Opus 4.6 + Figma Make.

Here are 9 insane prompts that generate complete websites in minutes 👇

Bookmark this https://t.co/1euA3evtpa

@mazeincoding Kimi master

@Kimi_Moonshot u win https://t.co/t7Y27LOIoM

if they dropped claude Opus 4.6 like 8 years ago I’d be studying literature history and theology instead of fuckass operating systems and databases management systems like god intended https://t.co/dHE0uWqMui

most of those people don't have a clue how to actually talk to an AI, much less talk to Gemini - and i'm including you in that.
Talk. stop prompting, start talking https://t.co/QjRzGEOWql

Most important thing that happened in opensource AI this week btw

Given that the Qwen team leads were forced to leave, there’s very low hope we get a paper on this which is quite unfortunate https://t.co/V7bIi95Egw

I repeat, one of the best benchmarks at this moment
You can see the frontier models going exponential on it
With a *very* firm lead of OpenAI
no Chinese lab except DeepSeek is in the running
Xiaomi even *regressed* to cut token use
StepFun will catch up to Speciale I think https://t.co/bGXkg6A50G https://t.co/W3uxnGxVD4

This is where it gets incredibly dumb. The framework panicked and tried to failover to every other model I had installed. It spit out a massive error log talking about missing API keys for DeepSeek, Minimax and hilariously, even demanded an API key for my LOCAL MLX model.

2/ Prototype with AI Studio (Gemini 3.1 Pro).

For frontend ideation it’s honestly the best model right now.

But here's the trick most people miss:
Don’t ask for the best UI.
Ask for the best UX.

Good UX = higher conversions.
UI can always be refined later.

Last 7 days trend