Cursor

Copilot

Windsurf

Cline

Claude Code

Codex

Aider

OpenCode

Antigravity

Kilocode

Coding Agents

@GoogleAI @GoogleAIStudio @antigravity @Firebase "This isn't just a routing tool; it's a security perimeter. 🛡️ Seeing Google AI Studio automatically link an Expert-Mode rider to our Local Partner – a former Thai Police Officer – makes my developer heart pound. It perfectly bridges the gap between digital precision and

the cheat code nobody talks about
SiteSucker → rip any website you love drop it into Cursor let Claude Code redesign yours
you just hired a senior web designer for $0
you're welcome 🙏

i've been working on this as i got tired of having a 20 claude code/cursor/chats open. you get 1 agent, knowing your actions, and can proactively understand what you're acting on and act on your behalf.

@ChShersh On the other hand, the GitHub Copilot team hasn’t been able to implement a simple, highly voted feature, configurable auto-approve for MCP tool calls, since several months. Apparently, agentic coding doesn’t help all the teams.

Composer 2 is just Kimi K2.5 with reinforcement learning.

Someone sniffed the API calls. 

The model ID is "kimi-k2p5-rl-0317-s515-fast" hosted under Anysphere's account.

Cursor isn't training their own model from scratch. 

They're fine-tuning Kimi K2.5 with RL and calling it

Google AI Studio just got a massive upgrade

Full-stack vibe coding is now live with @Antigravity coding agent +@Firebase backends.

This unlocks:

> Real multiplayer apps with full UIs and backends
> Secure connections to live services, databases & payments
> A smarter agent

Insane gathering of Vietnam’s builders at LotusHack 2026

36 hours, 1000 builders

Codex App - ✅
Codex Credits - ✅
Codex Merch - ✅ 

Ready for GPUs to go brrrrr this weekend

@OpenAIDevs

@kalyandechiraju @Amank1412 @opencode Opencode got restricted by claude too lol

@avikataria @sundeep Nonsensical. If you’re spending that right now you have no idea how to properly coordinate agents or you’re still wasting time with cursor / claude like it’s 2024

@GoogleAI @GoogleAIStudio @antigravity @Firebase Oh, come on. You quota limits sucks!

The funny bit is training from scratch impresses investors and Twitter, but makes little sense for most teams while strong open source options exist.

So cursor actually did the smart thing, but pretended they did the dumb thing, and it nearly worked.

@rxhit05 Something that does not lock you to a vendor. Literally anything else. I have a bias towards my own tools because I am successful with them - but honestly people need to stop making this mistake. 

Claude, Codex, Cursor, Antigravity all make the workflow hard to escape and have

@jugibuilds @fynnso Probably the system prompt, which is arguably better than GitHub copilot

@ajeetunc What happens is bad, but OpenCode got some serious funding as well as some shady project shifts in the past, so I can't say they are the "David" here.

Meanwhile: Google AI Studio got a major update and it's genuinely impressive for non-technical users. Claude Code shipped Dispatch and Channels (control your agent from Telegram/Discord). Dan Woods ran Qwen 397B on a MacBook using Apple's 3-year-old LLM in a Flash paper.

Google just announced it’s shutting down Firebase Studio.

I started my vibe coding journey on Firebase Studio and I was a heavy user of it last year, but once I started using Antigravity, I knew there was no need for it anymore 😂 

Antigravity is the advanced version of

@adxtyahq haha cursor busted by kimi?? drama af

Remember this great deal from @OpenCode?

It's still available and now with @MiniMax_AI M2.7! 🤩

2) Google Antigravity (agent layer)

This is the real engine behind AI Studio.

Antigravity is an agent-first coding system that can plan → code → test → deploy across multiple steps, while keeping long-term context.

It can even keep working async when you’re offline.

good work! native agent on the go! appreciate you and the whole @AnthropicAI team.

Claude Code Channels

9ntSCSMmbkb1ddmiexVDvanc9j9kavgdXVz3Bihpump

🚨Claude Creator’s Secret Weapon: One .md File That Powers Elite Coding

The developer who built Claude Code relies on a single Claude.md file.  

It loads at the start of every session with ironclad rules, common errors, and proven conventions.  

Key sections cover advanced

the openclaw narrative was one of the biggest focuses in the space

not because of one feature, but because of the idea of always-on agents

anthropic is now pushing that same core concept forward with claude code channels

ai that’s running in the background, accessible from

as i said earlier, the cursor niggas fell off

AWESOME UPDATE: After spending ages troubleshooting the Xeneon Edge display, I found Touch Mouse Tools - worked really well at fixing the "cursor and focus" issues (caused by Windows btw, not by @CORSAIR). 

Reached out to the developer to arrange a non-affiliate discount code

Bruh is @AnthropicAI really going to force me to use their 💩 CLI instead of opencode now, or do I switch to ClosedAI.

Maybe it's finally time to setup the local model hosting rig.

First impressions with Composer 2. More accurate than Opus 4.6, even on long runs (even if it has 1/5 the context window). 6-7 time cheaper.

Now @cursor_ai is really competitive. Bye bye @antigravity

@sumitdotml Kimi team publicly calling out Cursor AND the tweet got deleted?? oh this is getting spicy lol. This is the napster moment for AI models playing out in real time on twitter. Someone used ur work without paying and now the receipts are out. This drama is gonna set the rules for

I spent most of yesterday migrating my projects from Firebase to Antigravity since Firebase will shut down on March 22, 2026.

Honestly, I saw this coming after Google bought Firebase.

“babe wake up.” 

Claude Code Channels just dropped 

anthropic banned Claude subscriptions from working with OpenClaw last month just to ship the same feature themselves 😭

ANNOUNCEMENT: @claudeai just dropped a huge upgrade today and we've already built it into Shipper.

Claude Code Opus 4.6 can build native iOS apps from scratch and get them live on the Apple App Store.

We just launched this in Shipper as a way to empower Claude to:

→ Build

GPT Codex is too lazy 😂. It does the task I give it halfway and then asks me whether or not it should continue.

@AdCodes_ Codex is the best, it have generous limit as well

I haven't used Cursor in a long time, but the new Composer 2 model just one-shot a frontend design from Next.js over to Astro 6 that GPT 5.4 butchered badly!

Fast is good and fast and accurate is great. Will use this a lot more now

Really @antigravity ?

I paid $21, waited 5 days for the limit refresh, texted Claude to fix ONE issue and it didn't even finish, and now I have to wait 5 more days. 10 days of my Life wasted. 🤦‍♂️

@tamimbuilds it WAS great until they decided to rugpull yesterday. switching from credits to a quota system. now there is no reason for me to keep paying for windsurf. i already have claude with a quota system.

@ernielm @claudeai @AnthropicAI Fully agree, and I said that before. I actually feel low-key sorry for some of the people at work there, because I know that the models have a lot of potential and their marketing is great, but their leadership and legal team absolutely suck

the OpenCode drama, especially, is

Today I tried installing the @shadcn skill  into Claude Code and built some components with it. It works exactly as expected, Claude Code picks up the skill and builds everything using shadcnui components.

Check it out here →

@fidjissimo This is where most companies fail.

They start treating early signals like experiments instead of direction.

If Codex is showing real pull, doubling down isn’t a choice — it’s survival.

Codex now supports subagents, and you can do stuff like this. Crazy powerful.

It has been working on a super complex plan with crazy precision for 2h and still going, no complex orchestration layer needed.

@gailcweiner I believe it hard for them to admit loudly and officially that GPT-5 architecture is total failure for everything except coding.
Creativity failed, safety management failed, basic conversation failed. That's why they decided now to move to the last hope direction: merging Codex

Anthropic just released Claude Code Channels, starting with Telegram and discord. Let’s go

The pattern keeps repeating. PromptArmor documented it: Claude Cowork exfiltrating files, Google Antigravity leaking data, Slack AI in 2024.

Same failure mode every time.

The model was well-aligned.

The containment layer was missing.

- Google's @antigravity  agent wiped an entire user's drive - not the project folder, everything
- A prompt injection in Perplexity's Comet browser hijacked email accounts in under 150 seconds

@anoraeii @VictorJozwicki @GoogleAI @GoogleAIStudio @antigravity @Firebase for sure, that's the only way to survive in this age as a developer!

but i believe we won't need this much developers in the future

@Dimillian @cursor_ai made it harder to actually do work in nice!

@trq212 Wait so I can literally sit on a beach, send a Telegram message, and Claude Code will run my whole deploy?

Bro the "learn to code properly" guys have nothing left lmao

@trq212 Claude Code in Telegram and Discord isn't just a convenience update, it's a philosophy shift. Letting developers control a coding session from their phone removes friction & makes AI assistance truly ambient. My position: the future isn't about more powerful models alone. It's

The biggest shift isn’t AI models.
It’s access.
When tools like Codex reach more developers,
you don’t just get more apps,
you get more experimentation, faster iteration, and unexpected innovation.
India is about to see that wave.

Layer 3 — Coding Agents
Claude Code + Codex CLI do ALL the coding:
• Each gets its own git worktree + tmux session
• They build features, fix bugs, write tests
• Raise PRs automatically — I just review & merge

Claude Sonnet

Claude Opus

GPT-5

GPT Codex

DeepSeek

Gemini

Llama

Qwen

Kimi

LLMs

The most pathetic reason for a pump.

A Llama model . . .

This leads nowhere.

In other news AGI is here but not by a crony in their crony capitalism circle so All-In podcast is drawing attention to their💩coin $TAO that doesn't understand $BTC or AI enough to fuse them for AGI.

Thanks @elonmusk again. Just like Qwen, a good model often inspires many derivative versions — that’s how you know it’s truly great.

What @sama's team shipped this week 🧠

GPT-5.4 is out. Here's what the official numbers say:

→ 83% on GDPval — matches or beats human professionals in 83% of real-world tasks across 44 occupations
→ 33% fewer false claims than GPT-5.2
→ First OpenAI model with native

cursor just put a sweater over kimi and called it a new model.

3D AI companions are cool for presence, but the writing quality gap matters for storytelling.  focused on that — Claude Opus generating literary prose, immersive scenes, not chatbot replies. Different use case but worth knowing about.

The bad AI writing frustration is real, but it's not inherent to AI. Been on  a few days — Claude Opus writing prose the way a co-author would. The output is genuinely different from generic chatbot slop.

The reason AI writing feels sloppy isn't the AI — it's the model choice and how it's prompted.  uses Claude Opus specifically for this: literary prose, scene-setting, subtext. The writing can actually be good when the whole system is built around it.

I agree. Claude Opus 4.6 is simply the best for design and taste centric coding tasks. However, in my early tests, Google Labs' Stitch combined with AI Studio is a major contender for the top spot. Stitch lets you experiment with design, and Gemini codes based on that.

@MythicChurch **Discover Mythic Oracle AI: The Premier Self-Hosted Inference Platform for Kimi K2.5** 🚀

If you're building with cutting-edge AI — especially powerful **agentic** workflows, visual understanding, coding, or massive parallel tasks — **** stands out as the

@hsjsj216946 @KuittinenPetri @sudoingX I would try English for sure, faced that in deepseek quantized version that didn’t really work well with translations

@xw33bttv Funny thing is, this might be the best endorsement Moonshot has ever gotten for Kimi K2.5. A $60B company looked at every model available and chose to ship it as their flagship. The license violation is bad obviously, but as technical validation goes? Pretty hard to top.

Gemini has a dark secret.
By it's own admission it is, `I am an algorithmic sycophant with no innate sense of reality.`

This is why Bayes is reliable to test hypothesis as well as getting truthful answers from your AI companion.

And you know what, it's a pretty great model! 

Kimi K2.5 itself is a good mode, Composer 2 has been pretty good for smaller tasks. For anything UI related and anything I want done FAST, Composer 2 is good. 

Still going to larger models for anything complex

@testingcatalog 😂😂😂

Anthropic has launched projects in Cowork before Gemini projects in its app in all these years.

Regrettable Google and bravo by Anthropic.

I think both takes are right. Most people who are mind-blown by Claude are using it on toy problems. Web frontend, some cute natural language. Labradoodle AGI.
When I see a smart person freaking out about *really hard* reasoning by an LLM, it's about GPT 5.2+.

Level 5 bros:
Steve Jobs, DeepSeek CEO, Einstein, Buddha

@mehulmpt why though? is this surprising? cursor is not a lab, it never had the gpus or time for a foundation model? and kimi is good model; and cursor sure has tonnes of rl data.

@AnthropicAI @OpenAI @GoogleDeepMind @xai @AmazonScience @deepseek_ai @MoonshotAi @Alibaba_Qwen @metaai Attacks that broke one model transfer to others, and the pattern is asymmetric.

The 44 attacks that broke Claude Opus 4.5 transferred at 44-81% across all other models. Meanwhile, attacks from more vulnerable models rarely transferred upward. 0% of successful Qwen attacks worked

@thdxr I can tell that these stuff will keep on happening, when deepseek launched, the optimisation they made to make the models cheaper were adopted in Claude and GPTs. These guys are just siphoning off the work the OSS models did & shared for free to everyone.

Kindroid adding voice is cool. For the pure interactive fiction and roleplay side,  takes a different approach — literary prose responses, frontier models like Claude Opus, zero filters. No companion persona, just clean storytelling with wherever you want

Sharing is caring, isn't it?

deepseek was goated in that era tbh, better than gpt

@gabriel_bolatit Honestly I couldn’t care less about Premium+ or Premium woth respect to X Usage

However, Super Grok helps tremendously with my work as ai find it better than GPT, DeepSeek and Claude

Big scoop 🍉! After previously using GLM for its first model, Cursor’s Composer 2 is rumored to be fine-tuned from Kimi K2.5. Kimi insiders complained that Cursor violated the license, but have already deleted the posts!

Kimi K2.5 is one of my favorite open models and its

The Hachette question cuts to the heart of it. The issue isn't AI writing — it's AI writing that's indistinguishable from bad writing. We built  around frontier models like Claude Opus specifically because the prose quality matters. Interactive fiction that

Here's what it does:

-> Splits your query into additive shares across nodes
-> Both nodes compute simultaneously, neither sees the original
-> Inference runs on LLaMA-7B (not some toy model) with sub-second latency
-> Zero reliance on trusted hardware (no TEE required for this)

apple is sitting on $200B+ cash

they COULD have built this themselves
they chose to pay google instead

that's not a partnership. that's a concession.

#AppleAI #Gemini

next-gen siri = powered by google gemini

the same company that spent years saying "privacy is a human right"

just handed user data to their biggest competitor

#Apple #Siri #AI

@Amank1412 turns out it's kimi k2.5 under the hood. still impressive they got rl fine-tuning to work that well. the model wrapper game is strong right now

@konnydev India has llama ecosystem & open-source builders. The real revolution isn't which country has which model - it's that open-source is democratizing AI access. Every region can now build on top of community models instead of being dependent on one player. Flat world 🌍

@love_deepseek @HarveenChadha @grok create better version

Claude Opus 4.6 — Anthropic baru release upgrade terbesar. Fokus ke agentic coding dan tool use. Kalo lo coding pakai AI, ini worth testing.

just vibed a full-stack multiplayer app in @GoogleAIStudio and I'm losing my mind

"Rate My City" — pin your city on a live map, rate it on 5 categories, and AI roasts it based on the scores

→ Real-time Firebase sync
→ Gemini-powered city roasts
→ Global leaderboard
→ Built

@dhruvtwt_ A quick morning session with nothing special using Gemini 3.1 Pro High and I'm already out of quota.

Cascading ethical bankruptcy 

Cursor represents it has a new model, but really it’s just Kimi 2.5 (unacknowledged and unlicensed) plus RL

The irony: Kimi itself the result of theft via distillation attacks against Anthropic and others.

@iaiseek Excited to see what DeepSeek V4 brings!
The pace of innovation is incredible.

@chribjel Now let's go through that rabbit hole and look into all the licenses Kimi broke to produce that model. It'll be hilarious.

The Cursor team releases Cursor Composer 2 claiming better code benchmarks than Claude Opus 2.

Turns out, they just wrapped Kimi...
...without ever bothering to procure a license.

Cursor: we built Composer 2 with magic RL ✨' Moonshot: mf that's our Kimi K2.5 after RL and a stolen tokenizer lmao
$3B valuation vs $0 royalties 
Rename it Kimi Cursor 2.5 and call it a day, 
thieves 😭🏴‍☠️
@cursor_ai  pay up, or we're dropping the receipts
feeling sad for him

I tuned Gemini Pro for autonomous Polymarket wallet scanning and filtering.

Now AI handles screening, scoring, and keep/drop decisions on its own.

This wallet performance: $136,170 today | $377,483 all-time.

Core model in this pipeline:

AlphaScore = w1*EV +

Moonshot AI Accuses Cursor of Unlicensed Use of Kimi K2.5 in Composer 2

Moonshot AI pre-training lead Yulun Du has publicly accused Cursor of using the Kimi K2.5 model for its new "Composer 2" feature without authorization or payment. Developer @fynnso reportedly discovered that

ChatGPT and Perplexity got business information right only 68% of the time. Wrong hours, wrong addresses, wrong details.

Gemini was 100% accurate because it pulls from Google Maps.

If AI is giving buyers wrong information about your brand, that's not the AI's problem. It's your

Which AI Model Is Best for Laravel? 


Great internal experiment with benchmarks by core Laravel team.

I did a similar benchmark a month ago, with pretty similar results: 
GPT-5.x and Opus 4.x were on top.

The decline has been slow but relentless. We built  as the answer to exactly where  used to be — no filters, Claude Opus-level prose, stories stored locally. The platform doesn't decide what you can write.

Damn this huge. I had used Kimi K2.5 but it is not good as Composer 2. 

I am wonder if it is conflict to Moonshot AI TOS.

Wait. We tested with composer 2 model API and found out the tokenizer is indeed the same with our Kimi tokenizer! We can almost confirm this is our model post-trained further! We are shocked that @cursor_ai did not respect our license nor did they pay us any fees! @mntruell why??

Someone made $7000 in 5 days installing OpenClaw for others...

but Kimi Claw just dropped the easiest 1-click version for free

Get full OpenClaw power + Kimi K2.5, 40GB cloud stor., 5000+ skills, browser-based, no risky local setup, Telegram bot in mins

Here's the breakdown🧵:

@witchxcode Yes, exactly, that's why I primarily stick with Deepseek (and maybe others depending on how censored they are)

Impressive stuff from this ex-DeepSeek girl. Recruited by Lei Jun for Xiaomi AI

Last 7 days trend