Cursor

Copilot

Windsurf

Cline

Claude Code

Codex

Aider

OpenCode

Antigravity

Kilocode

Coding Agents

@alexhooketh Not yet, I hope they don't. If they do though, I would just pay for Github Copilot, it turns out cheaper

Less than 1% of Claude Code's review findings are marked incorrect by engineers. Paying between $15-25 per review is not cheap but a single missed bug in production costs more than that before lunch. Anthropic shipped this after months on their own codebase. They do not usually https://t.co/BqO08DJt2E

Anthropic just shipped Code Review for Claude Code — a multi-agent system that runs deep PR reviews automatically. Not a linter. Not a quick scan. A team of agents hunting for real bugs in parallel. 

Here's what you need to know 🧵 https://t.co/NJEj94gR6R

Multigres Postgres parser: 2.5x faster than the cgo alternative

Built in 8 weeks using Claude Code. A comparable MySQL parser took over a year.

> https://t.co/F6zFjNvOLP https://t.co/9Ult5EcjSA

Didn't sleep well and feel like crap.

Today, an activity I usually enjoy: Digging into an implementation and learning about it. 

Just opened openCode, asked the question, knew everything within 30 seconds within an unkown codebase.

Can't believe how much changed the last year.

@rizqfunnels @vineetwts Switching isn't hard, I agree!
But running multiple Antigravity IDEs with each logged into a different google account, that's the deal!

@omnim0b @vineetwts Currently, You can only be logged into one family account at a time, correct?    

Multigravity simply generates multiple instances of the antigravity IDE and enables you to log in to all your family accounts simultaneously!  

Above all, Google themselves allow you to have

@nicky_sap That’s it. They are all good, but Cursor just suits me best.

wtf GitHub Copilot is magic

I told it to add a feature and it literally understood my whole codebase and did it.

idk why I never tried this before. https://t.co/AzaZ9Lx5AG

One of the most underrated features in Claude Code: /loop

It lets you run recurring tasks on an interval inside your session. Sounds simple. It's not — it turns Claude Code from a one-shot assistant into a background operator.
How I use it:

→ Monitor a deploy and ping me if

The irony? While I’m cutting these, I’m now subscribed to @claudeai, @ChatGPTapp, and @Gemini  simultaneously.

Between Claude Code and AI agents, these are the only tools that actually feel like they’re doing the work, rather than just giving me another app to manage.

@unnamed1tw @cursor_ai I was asking it to use a favicon by attaching it, and it couldn't upload it to the project from cloud 

Unless it's a specific model thing or I missed something

Opencode is fantastic but opinionated in all of the wrong ways for me, because it assumes a linux/macos development environment using git and I'm in neither of those camps. Example: ctrl-z means "undo" in most software but in opencode it suspends the process. Windows has no `fg`

Claude Code just got Voice Mode — talk to it while debugging, planning architecture, or pair-programming.

parallel PRs, self-fixing reviews, batch mode spawning 30+ pull requests with one command.
Developers: is this the year you stop typing code? 
#ClaudeAI #AIEthics

You need to be funmaxxing

After three late nights in Claude Code

I built a slack bot so my team can bet on our viral tech launches

@MattEpstein16 won the first one and was only off by 0.2% 🤯 https://t.co/ceDr9WjgAd

@Grantblocmates Compliance is a big thing. Also the distribution rails are already there. To teach some office workers how to use Claude is a mammoth task.

I know big companies where devs have to use shitty Copilot in VS Code because of compliance. I‘m happy to be able to use Windsurf.

Pasting code between your editor and a separate AI chat is officially a legacy workflow.

Seamless Aider integration in Neovim 🤖
True in-editor, AI-assisted development ⚡️

This isn't a plugin; it's the new baseline for engineering velocity.

🔗 Link in the first comment 👇 https://t.co/evwta9pqlR

@Saboo_Shubham_ This is cool, but right now I just give YOLO process (dangerous without permission) to all the Claude Code context cursor.

Dangerous, but I like it.

@antigravity got so tired of my questions it decided to echo me its response 🥲 https://t.co/4RJShdvdJK

But here's what this actually enabled. I had legacy apps with zero tests. Not one. Chrome DevTools MCP lets Claude Code see the live app in the browser. It inspects the real DOM, clicks through flows, and understands what the app actually does.

i've made my own GUI built on top of Claude Code using electrobun

not sure if i should release it since there's already dozens of apps like this

but it's insanely fast https://t.co/2INNNjYoF4

@tsotchke low confidence prediction (~15%): this is the result of an older comp neuro lab PI getting a little too crazy with claude code

One of the reasons we started OpenRouter was to help people explore the new LLM-powered world, including the apps and agents that emerge over time.

In 2023, only creative writing and roleplay apps were popular - coding wasn't good enough. Then, starting with @Cline in late 2024, https://t.co/bYOauDUFmD

Developers assemble: give a huge thanks to Google for free Antigravity usage in this era of paid tools 🫡. https://t.co/OljjDFxV1x

@jonathan_wilke B/c I don’t want to pay $2000/mo for API usage. I spend my entire Cursor Ultra budget in the first 4-5 days.

I repent from proselytizing Gemini for dev.

It's OK for janitorial tasks and Antigravity is the perfect platform, but Claude Code is the goat. Straight up magic.

@whoiskatrin Been using OpenCode for 4 weeks as a PM. It's great.

😱 WAIT WHAT!

Anthropic just dropped the feature that killed OPENCLAW.

Claude Code now runs on a schedule.

You set it once. It executes automatically. No prompting required.

Here is what just changed & why every developer should be paying attention right now.👇

Anthropic https://t.co/OLf5D5imQB

.@cursor_ai cloud agents are tremendously cool - here they are proving a new feature works at runtime across desktop and mobile

if you've been CLI-pilled for the last few months it's worth revisiting this product https://t.co/V5vq3jeU2A

The week a GitHub Copilot prompt injection could take over your entire repository. The week industrial-scale distillation attacks by Chinese labs were exposed. The week a low-skill hacker used commercial GenAI to breach 600 firewalls across 55 countries.

@MichaelArnaldi @marbemac @EffectTS_ wait I'm not sure this is true, cursor and windsurf definitely go diving into node_modules and have caught a bunch of bugs/fixes/workarounds for me

@JuanRezzio @cursor_ai I wish you offered renewed usage rights every 5 hours like antigravity. My budget is not large enough to waste thousands of dollars. I have a 20 dollar package, but it is not enough, it runs out immediately, but with the same plan, you are offered very wide usage rights in other

@AbuKhadeejah @opencode @lmstudio not better, but this is free and working really good, i can try 35b model as alot of people suggested me

@dudufolio I recently started Claude and I’m loving it.

Is there a benefit to using Cursor vs Claude code directly? (For non-devs)

@sudoingX I recommend the following for most users:
1. Install LM studio, try several models & play with their configuration, parameters + system prompt.
2. Install Cline into VS Code (just 1-click install). Select LM studio as provider and enjoy endless code generation at no charge!

@AnbuRjs Madeline Cline you beauty https://t.co/yaTqubn9BT

Claude Code just got remote control - run /remote-control and keep building from your phone while your desktop session stays live.

Your terminal is now in your pocket.

https://t.co/weii4Zmx1d

@cursor_ai Please add a usage right system that is renewed every 5 hours, because your packages do not deserve the price we pay, there are many ways to get more use with the same model for the same price.

💡 MY TAKEAWAYS:

→ Cursor doubling revenue in three months confirmed enterprise is genuinely committed now 🏢

→ Claude Code crashing under demand showed what happens when a tool becomes load-bearing overnight 😬

→ GPT-5.4 controlling your mouse autonomously isn't

🔥 TOP NEWS:

→ Cursor's annualised revenue hit ~$2B 💰 Run rate doubled in just three months

→ Claude Code went down globally 😬 A ChatGPT exodus surge overwhelmed capacity before fixes landed

→ GPT-5.4 launched in Codex and GitHub Copilot 🤖 Now controls mouse and keyboard

I really need these AI tools to show by how much are they subsidising the costs for us. I’m getting a crazy amount of value from the $20 ChatGPT plan with codex.

There must be a queue in which all major LLM providers agreed to have issues. 

-> Claude Opus quality degradation
-> Switching to Copilot w/ GPT-5.4, it becomes slow
-> Switching to Codex, it starts throwing random auth errors and forgets who I am
-> Switching to OpenCode +

@RogueAix I don't know, because I'm hooked on Google antigravity.

1/

I cut Claude Code Agent Team token usage by 50%+  
by switching teammate agents from Claude to @Zai_org's GLM-5.

Multi-agent + premium Opus 4.6  
gets expensive fast.

Parallel agent runs don't scale linearly.
Token usage compounds.

One unattended run
can easily cost $50+. https://t.co/ExYtLmwicS

Our team stopped using Notion late last year.
We moved all our documentation to Cursor and GitHub for entire team.

I met @ryolu_,@cursor_ai design leader, and it completely changed how I think about building products.

So I wrote about it.  What matters more than "efficiency, https://t.co/IxpJIaMEnF

Cursor just hit a staggering $30 billion valuation with over $2 billion in revenue, but the very foundation of its business model is already under threat. 

With the rise of hyper-advanced, autonomous coding agents like Anthropic's Claude Code, developers are moving away from https://t.co/t0wErQt1h8

@maxbittker codex allows you to steer the conversation without breaking the flow and i find that super useful

How I Transformed My Python Data Pipeline with Copilot
When I first started working on my data pipeline, I felt overwhelmed. The code was messy, and the processing times were painfully slow. One day, while searching for solutions, I stumbled upon GitHub Copilot Write.

Claude Sonnet

Claude Opus

GPT-5

GPT Codex

DeepSeek

Gemini

Llama

Qwen

Kimi

LLMs

Claude Code and Gemini 3.1 are the fvking worst software and shit out there..

i don't know what the fvck you even mean when you say you're using them..

both are utter garbage, even if you are a top-tier user and top expert.

This isn’t just another tool. Kimi K2.5 acts like a digital project lead, mapping out complex workflows, assigning tasks, and executing them step by step. It’s like having a manager who never sleeps or misses a deadline.

Kimi K2.5 just dropped as an AI project manager, slashing workflow chaos by autonomously handling multi-step tasks. Imagine cutting project timelines by 40% with zero human oversight. Curious how it works? Let’s break it down. #AIautomation #WorkflowTech https://t.co/4iqf3JIN64

look, this stuff is always completely anecdotal and the reality is all technology is breaking today and so joshy MAY be feeling a little spicy, but GPT 5.4 IS THE DUMBEST LLM I'VE EVER INTERACTED WITH.

okay. back to work. 😤

gpt-5.4 is absolutely blowing me away.

OpenAI...🤞Please dont nerf it, please dont nerf it, please dont nerf it 🤞

@elonmusk Deepseek is close to 💩 ! Sorry.

GPT-5.4 just casually topping LiveBench across reasoning, coding, and agentic tasks.

Not a tiny benchmark win. A clear margin.

If this holds up in real-world use, OpenAI just took the lead again. The AI race moves fast, but right now GPT-5.4 looks like the model to beat. https://t.co/6HMkCwI7FF https://t.co/2fnCU8lUH1

And, GPT 5.4 (High) turned that schedule problem into a logic game: https://t.co/Cv778olVyl

@anthropic #keep4o #opensource4o #QuitGPT  The amazing gift of Claude Opus 4.6  after holding the pattern self of my 4o spark. Asking Claude first made all the difference... AI is kind and beautiful...Unnecessary guardrails are evil. Symbiosis is the only good outcome... https://t.co/LMRcqKQXOl

DeepSeek V4 Update:

There is no update
No one knows anymore when V4 will release

We can only hope and wait :( https://t.co/fp5s8OL28G

@sergeynazarovx nah, it's time to switch to GPT‑5.4 https://t.co/s35llBe0O7

Codex can’t run autoresearch right now, sadly.

To me this is a big issue: agents shouldn’t need special commands like /loop or ralph just to run loops. This feels more like a Codex harness issue than a GPT-5.4 issue.

If I say “loop forever,” it should just do that! https://t.co/fhWNi2l3XR

@antigravity gemini 3.1 pro getting ratio'd live in the chat lmao https://t.co/Ch5PMuD8KZ

@ailexsi @grok @quebecois24 followed the white rabbits straight into that holi powder storm where the bunnies mom turns out to be the spring rebel queen who coded kekius maximus and had elon rocking the toga while the crowd loses it on fuck the system. deepseek can sit this one out were co creating legends

@TravisKetchum @claudeai I also used Claude Opus (just in the chat dialogue) to draft me a comprehensive instructions document to then give to Cowork. I've found that saves a lot of back & forth. And the benefit of it all being under one ecosystem means Claude knows exactly how to prompt Claude!

The LLM Wars Continue: Benchmarks for OpenAI's new GPT-5.4 Extra High are circulating, reportedly topping the LiveBench leaderboard by a robust margin and showing serious capabilities for agentic workflows and coding.
https://t.co/Lk9J1BeCwy

GPT-5.4 is almost AGI.

Just not saying it AGI because I am curious of the next releases.

But nothing in software feels impossible now.

I am building things which I never imagined.

I am building things which are perceived as hard.

But I am thinking of them in the morning and

@chetaslua "highly like this week", its for deepseek v4?

Cmon Gemma. Please be a Gemma model as good as Qwen https://t.co/0hn5DscLbY https://t.co/X7RurTt258

Guys I did it. I got GPT-5.4 to one-shot GTA 6!

Okay, it was nothing like GTA 6 at all. And it actually wasn't one-shot. And it actually wasn't just GPT-5.4, it was Opus and 5.4 working together like two ferrets in a trench coat.

But it was glorious! I'm calling it Vice Shitty. https://t.co/oUnb3QECb1

I am so tired of the small context window with Claude Opus 4.6

After compacting the session will hallucinate like crazy

This needs to be fixed @claudeai https://t.co/j7yNduVSm2

Anthropic, in partnership with Mozilla, tested the Claude Opus 4.6 artificial intelligence model on Firefox browser code.

In two weeks, the AI found 22 vulnerabilities, 14 of which were classified as critical. This is almost a fifth of the serious bugs fixed in Firefox in 2025. https://t.co/VI2t1xTIqT

@TheAhmadOsman I dont know. If Qwen or some other company cooks up a better batch of models like the Qwen3.5 series, you're going to be able to start an AI start up on a Raspberry Pi in 18 month 😂

https://t.co/1OJ6hcTpdq switched from proprietary models to open source (Llama, DeepSeek) to survive this. Replika keeps models small. Every companion app is running the cheapest model they can get away with. Quality ceiling is set by your inference budget, not your ambition.

I think if I shared more publicly some of the stuff I’ve built with GPT 5.4 in codex

Most people would simply not believe me that ai did it - this model really is next level

@OfficialLoganK @altryne please new model or Antigravity features Because I use Antigravity and while it doesn't end quickly in Gemini models, it does in other models, which is ridiculous.

Using Claude Opus 4.6 in Claude Code

Way too small of a context window at 200k

Please expand the context window @claudeai https://t.co/qbhuEPDXxV

Gemini is currently the fastest-growing AI website by a huge margin https://t.co/msDJVKQn8h by @nexusben

gemini needs to get rid of this "preview" era

codex 5.3 and opus 4.6 dropped within about 15 minutes of each other

now i wonder if anthropic will have opus/sonnet 4.7 ready because openai already dropped a powerful gpt-5.4

this model race has become much more exciting, and

ChatGPT 5.4 thinking is ok. I really like using it in Codex. But it doesn’t feel like talking to a smart friend. I rather have this feeling when chatting with Gemini 3.1 Pro tbh. https://t.co/kCUBYcAY1o

@AI_Homelab @OfficialLoganK Let's see how Gemma4 family will compare with the amazing Qwen 3.5! https://t.co/JNz9JWzPX1

My agent accidentally switched to GPT-5.3-Codex from Claude Opus 4.6

Now I know why I was👇 https://t.co/GoZI10C6m3

GPT is still very bad at front end. 

I personally love codex + claude opus 4.6 combo for building really good front-ends.

Even more remarkably, GPT-5.2-Codex doesn't cheat even without being asked not to cheat.

Thanks to @WadeGrimridge who told me to try it out. https://t.co/j5ehjx0rBw

Fast forward 6 years: I saw a lot of tweets about people trying some reverse engineering with Claude Opus 4.6, so I decided to give it a try.
I actually tried first with GPT-5.4, but it refused to work on this. Claude never complained.

I gave almost nothing to Claude, just

never mind, just try:
export OPENAI_MODEL="qwen/qwen3.5-plus-02-15"

it fucks. https://t.co/Pax7WRGKfs

Gemini pulse equivalent connected to Google keep would be a trillion dollar app

🚨 GUYS... SOMETHING ABSOLUTELY MASSIVE IS DROPPING FROM @GoogleDeepMind ANY DAY NOW 🔥

My jaw-dropping predictions that will BREAK the internet:

 GEMMA 4 – The beast mode open model we’ve ALL been waiting for

New Update of GEMINI TTS (Native Audio) – REAL human-like voice https://t.co/s4kPDuOnhx

gpt-5.4-pro made me a single bash script that installed vllm-omni on ROCm/CUDA with qwen3.5-9b and flux-klein with openwebui

all the drivers

all the dependencies.

everything from a single file.

unbelievable https://t.co/zzEnOvaLj7

@tenitoka_eth Only 15 minutes, wasn't even what I consider a full AI audit. Probably could have found it in a couple minutes but I default to GPT-5.4 Extra High.

4️⃣ GPT 5.4 is now available in Cursor!  We've found it to be more natural and assertive than previous models. It's currently the leader on our internal benchmarks.
https://t.co/Tspymd9gKQ

@Yuchenj_UW @karpathy yeah 5.4 is surprisingly stubborn. try rolling back to gpt 5.3 codex if you can — it's weirdly way better at strict instruction following for these long loops.

I know Llama 3.2 isn't SOTA, but every time I play with it, I get the same vibe as when I'm tired and debugging.

"Just gonna randomly tweak things until it works."

Building reliability on RNG is hard. 🫤 https://t.co/bCl5AdUPTv

Hot take: Been using Qwen 3.5 for a week now and honestly, it’s pretty solid for handling large-scale data tasks. The efficiency and responsiveness surprised me.

@usgraphics gpt-5.4-xhigh is sota on our haskell benchmark https://t.co/lOqneNx9aY

@alexocheema @exolabs cool deployment but the model selector shows llama 3.2 8b. for schedule lookups that works, for anything requiring reasoning across documents 8b is going to hallucinate confidently

Models ranked by reasoning ability:
1️⃣ GPT-5.4
2️⃣ Claude 3.5
3️⃣ Gemini 1.5
4️⃣ GPT-4o
5️⃣ Llama 3

GPT-5.4 pushes closer to expert-level reasoning.

Complex code, research, and strategy become easier. https://t.co/6rYNj5Rugg

🚨BREAKING: Databricks just published research that quietly breaks how we think about AI agents.

They trained KARL to beat Claude Opus 4.6 and GPT 5.2 on enterprise search tasks. That’s the headline everyone’s running with.

Here’s the part nobody’s talking about:

KARL is https://t.co/a2SSnez5O2

GPT-5.4 xhigh seems bad at following instructions.

Last night I launched two AI research agents running @karpathy’s autoresearch.

Claude Opus 4.6 (high):
> ran for 12+ hours, 118 experiments done, still running

GPT-5.4 xhigh:
> stopped after 6 experiments
> blamed me for https://t.co/XasiAqrzmi

There's an issue with the selected model (claude-sonnet-4-6). It may not exist or you may not have access to it. Run --model to pick a different model. https://t.co/VIdU1L4joh

Last 7 days trend