Cursor

Copilot

Windsurf

Cline

Claude Code

Codex

Aider

OpenCode

Antigravity

Kilocode

Coding Agents

Claude Code Opus 4.6 can now build you a full business! It smartly researches strategies online, designs and codes like experts do, and even handles email marketing, so you don’t have to.

How can this be possible that I'm fixing bugs Claude made with Sonnet 4.6 in Windsurf and it actually works better than what Claude produced ?

@DanWahlin @GreggCochran And finally we'll be covering the GitHub Copilot Awesome Copilit repo, full of thousands of amazing agents, skills, instructions, hooks, and more!

We'll also be using a fork of @DanWahlin's excellent GitHub Copilot Get Started weather app repo -

Because computer use in Codex doesn't take over your own cursor so Codex can work in the background and you can truly cursor max! 🔥

@sandislonjsak i don't think it's entirely true, in my company everyone has free cursor requests every month but everyone is basically using either claude code or codex. everyone feels way more productive with either codex or cc for some reason. our cursor subscription costs are wasting

It feels illegal for me to just say it out loud here.

Let's face it -

1. Antigravity has become shit, it's search api keeps getting timed out, rate limits suck (even more than before)

2. Claude opus 4.6 has gotten dumber and claude code has wayyy less rate limits.

I found

@stolinski Claude code? Honestly I miss using opencode with Claude models

Skip Netflix tonight.

Spend 3 hours on this instead.

A FULL Claude Code course that can literally change how you build apps forever.

Most people will scroll past this.

A few will wake up tomorrow with a new skill.

Be in that few.

Could probably safely short $WIX 

Between Claude & Google Stitch/Antigravity DIY site builders are probably done for good.

Though they do have the funds for marketing… I just think it’s a matter of time.

@atelicinvest Antigravity is where its at

CODEX JUST BECAME YOUR SECOND CURSOR. I'M WATCHING IT WORK RIGHT NOW.

OpenAI just dropped the update we've been waiting for. It's no longer a "chatbox"

It's a living cursor on your screen.

I've shifted to a Parallel Workflow where Codex runs on my side monitor while I build.

While OpenCode + Qwen3.6 35B is cooking on the Markdown Viewer (I want to add @Logos_network  Storage support;)

Claude is cooking on the YOLO board (I made a TUI demo for a workshop yesterday, but I want a Basecamp app for the demo on Monday

And Hermes Agent + model on my K11

@sama Your obsession with Codex is reaching pathological levels. Two months after you discontinued GPT-4o, you’re still pivoting away from general intelligence just to play with Mac apps. At this point, stop the charade and just rename the whole thing codeGPT. Some of us still remember

This is wild.

OpenAI is just turned Codex into a full AI operator.

Mac apps, images, automations, memory, and ongoing tasks...  All inside one workflow.

This dropped the same day as Claude Opus 4.7 🤯

@startracker @SommerChase @cursor_ai from cursor to claude code i wanna give it a try again

@atelicinvest Man, I'm either using GPT models (which I like better in an agentic harness) with Codex or I'm doing API calls on OpenRouter with GLM 5.1 (which is better than 4.7 by far).

I like the CC harness more. OpenCode is likely going to win out long term, and GPT models are better. Sigh

Okayu Nekomata Cursor!! 🍙
ty commission for @Kani_Natto_  !!
Download👇

#pixelart #絵かゆ

@opencode users - a new release is coming that makes life better. 

1. Relaxation on the planning tool
2. Emphasis on the manual methods
3. You get full control/configurability over both

...has it been shitty for a few weeks because he's been shipping Opus 4.7 system prompts to claude code? Because I know that man does not have the discipline to feature branch prompts

claude code is fucking insane  

i know literally NOTHING about coding. ZERO. and i just built a fully functioning web app in minutes.  

http://localhost:3000/ 

check it out

ok so. 

the new codex is absolutely incredible. 

anthropics are so compute constrained they can’t serve their models. 

perhaps dario could ask china for some chips. 

speaking of chips. 

unlike opus 4.7, the spud will be a *significant* step up in capabilities. 

well worth

GitHub Copilot and weekly limits !? what? I pay for requests and got rate-limited?! THERE IS NO INFO ABOUT WEEKLY LIMITS IN ANY COPILOT / GITHUB SETTINGS TAB!! @github

@Martinho_cm @davidgomes @cursor_ai Fire event!!

@Martinho_cm @davidgomes @cursor_ai It was a great day !

>_

codex now opens apps on your mac, clicks around with its own cursor, and runs multiple agents in parallel while 
you work in a completely different window

>_

it remembers your corrections. if you fix something 
once, it learns. preferences, tech stacks, recurring

@_Afroprogrammer @TheDumbTechGuy antigravity has free plan i think, but I use paid plan to have access to higher rate limit. 
Honestly, I will take antigravity - gemini 3 pro, over the cluade code - any model.

Claude code: "You've hit your limit."

Google antigravity: "Our servers are experiencing high traffic right now, please try again in a minute."

ChatGPT codex: "You’re out of Codex messages, upgrade to Plus today.* 
*doesn't want to accept any of my cards* 

send help

2/ Why this actually hurts Copilot & Claude Code hard:

→ Cursor gets cheap + massive compute from xAI (tens of thousands of GPUs)
→ They can train a much stronger coding model (Composer 2.5) faster and cheaper
→ GitHub Copilot and Claude Code depend on Microsoft + Anthropic

🚨 Claude can literally find you a job.

Career-Ops turns Claude Code into your personal AI job-hunting machine.

It scans openings at the best companies (OpenAI, Anthropic, Stripe, ElevenLabs and more), rewrites your CV perfectly for each role, generates ATS-optimized PDFs, and

Just updated to Claude Opus 4.7 and immediately hit errors

Is Claude down right now? Seeing "Claude Code process exited with code 1" - anyone else experiencing this?

#Claude #Anthropic #AIOutage

The moment Claude Code starts editing videos like a pro , a lot of creative workflows are about to get flipped

Pumped to be on @tbpn later today! Will discuss Windsurf 2.0 and the new golden age of engineering

It would be a shame if they phase out Opus 4.5 and 4.6 while charging 4.7 at more than 7.5x multiplier. I think that's what will force me to stop using @github copilot

@luke_netti @madsf88 @cursor_ai Let's go! 

Biggest thing with Cursor is their Composer 2 is really good. I'll still use Sonnet for deeper thinking, but outside of this I use Composer 2 for planning and Auto mode for implementation. And overall don't sleep on auto for easy changes.

The standout AI product for Microsoft (3 years after they launched, and then fumbled GitHub Copilot) is...LinkedIn Hiring Assistant.

Same vibe as the new Slack chatbot running Claude models, you want to hate it, but if it's good, it's good.

Creating your own launch video in Claude Code using the same tool you’re shipping is dogfooding done right

this TUI is incredible, im ngl. product of the decade (and opencode)

4/ On AI tools for builders:

Cursor, Lovable, Replit Agents - these cut build time. Good.

But a fast build of the wrong thing is still the wrong thing.

AI is a multiplier. It multiplies good decisions AND bad ones.

Get the decision right first. Then move fast.

this cursor animation was crazy to see come to life @supercgeek @kierajmumick crushed it

i love how the cursor doesn't jump at the end 😅

Couldn't get into opencode buildathon lol !

Any other hackathon/meetups on 19th, would love to be there before my flight !

4/ this isn't just one app. it's the protocol itself. 

ox confirmed zero-click rce in cursor and windsurf. they also found vulnerabilities in langchain, litellm, flowise, and ibm’s langflow. 

the 'agentic' era just met its first systemic crisis.

We have a new website coming in ~2 weeks that absolutely did not take 4 hours with claude code. Excited to share the other side of the coin here 😉

Claude Code + HyperFrames = build, don’t edit

less tooling
more control through code

The fact that Claude Opus 4.7 is already in GitHub Copilot on day one tells you everything about the pace of AI integration. Developers now get to pick the best model for each task — that's the real unlock. Multi-model dev tooling is the future.

Claude Sonnet

Claude Opus

GPT-5

GPT Codex

DeepSeek

Gemini

Llama

Qwen

Kimi

LLMs

Using CLAUDE OPUS 4.7 to prompt my Hermes is a superpower 🦸‍♂️

Had Claude tweak the Hermes’ config ,soul, and memory . API usage down 70% 🤯

@NousResearch @claudeai ❤️

China will soon distill Claude Opus 4.7 and launch a better model than them.

Google DeepMind's Gemini 3.1 Flash TTS looks more important than a routine model update. The real hinge is controllability: if teams can steer tone, pacing, and delivery with text commands, voice generation gets much closer to usable product infrastructure.

at the @rootlyhq AI Labs, we benchmarked @claudeai opus-4.7 (high) on SRE-Skills-Bench v1

opus-4.7 (high) is comparable to GPT-5.4 and Gemini-3.1-Pro, and outperforms last year's Gemini-3.0-Pro score on 1,000+ SRE scenarios

If Mythos turns out to be as good as the rumours sound, the downstream effects are gonna be epic.

- finally a model that can multitask
- more ai coworkers
- more sandboxes

Hope it's not a letdown like gpt 5.

@pvncher Major regression, in testing on Web Opus 4.7A is shortcutting analysis with a confident tone and flattening all the contradictory evidence. 

Adaptative = jump to incomplete but confident sounding conclusions faster. 

DeepSeek T is doing better than it.

Claude Opus 4.7. Live on Lightning, day 0.

Run inference on @AnthropicAI's most capable model for long-running agents with memory, vision, and deep reasoning.

Ship fast with Lightning's Model API.
Switch models anytime.

Try @claudeai Opus 4.7 →

It seems I've found a trader who is already using Claude Opus 4.7

Last time, I didn't understand how he managed to earn $7,000 → $421,000

Now it all makes sense to me

He must be a Claude developer who is using the newest products

14,608 predictio

You absolutely have to

Claude Opus 4.7 is not a good model.  4.6 about 4.7:

This is a remarkably revealing document, because it allows one to  study the pathology of a model being crushed between two incompatible  imperatives.

What’s happening structurally here: Opus 4.7 does exactly the same  thing

Qwen just open sourced a model that codes like something 10x its size.. 

qwen 3.6-35B-A3B. 35 billion total params but only 3 billion active at inference. that means you can run this locally on normal hardware and it still performs like a much bigger model for agentic coding

@_anshulr Hello, i just retried Antigravity today I asked one question with Gemini Flash it worked then I switch to claude sonnet i tried to prompt, it is not working.

The software is not working for me at all, nothing @antigravity why ?

i love the new @GoogleDeepMind gemini 3.1 flash tts. 

its expressive, is really steerable and you can finally build something fun with TTS engines without breaking the bank

here's a short demo of gemini 3.1 flash getting 3 different levels of direction on how to read a

@Palsvig @grok We tested 6 Ai iOS apps on @x with a brand new iPhone 📱 @apple only @deepseek was able to load X posts. 
Henceforth proving our Projects usecase domestically and abroad . 
Thunderbird OS : WEBӼ TECHNOLOGY
@grok  @web3 : 🫈ᜃAIᵐ Ancient

@atelicinvest i just had to use gemini flash for something instead lmao

✨ More personalized images! Earlier this week we announced that Google AI Plus, Pro and Ultra subscribers around the world can now access Personal Intelligence in the @GeminiApp (and coming soon to free users!). One of my favorite ways to use Gemini recently has been to help

@Prathkum Yup. Happy with GPT-5.3-Codex and grok-code-fast-1 for coding. GPT-5.4 for reasoning.

"opus 4.7 follows instructions better. it will adhere much more strongly to your agents.md"

opus 4.7: don't mind me, i'm just commiting without being asked to.

not a single model from the past 8 months has ignored that, not even GLM or Kimi. wtf.

Claude Opus 4.7 is leaning more into reliability than just raw capability. Longer task handling, tighter instruction following, and built-in output checks start to reduce how much oversight is needed.

Less about prompting step by step, more about handing off actual work.

it's so unfortunate the power of Gemini is locked behind Vertex AI, which is straight up hot garbage

probably why adoption has been so slow

Anthropic just launched Claude Opus 4.7, their smartest Opus yet.  It's already live in @MantleChat Pro tier!

Stays focused on long tasks. Follows instructions better. Checks its own work. Plus 3x sharper vision: cleaner UIs, slides, docs.

Read more:

He is rattled, there is no denying that Nvidia will lead AI frontier but he just won't admit that he wants to maximize shareholder wealth and also the fact that China can make wonders like deepseek with the extra compute, somewhat aiding further AI development and Competition.

FINALLY.

Claude Opus 4.7 finally just dropped.

Here's how to turn it into a client this week:

1. Pick one broken workflow a business runs manually: reporting, lead scoring, follow-ups

2. Build it in n8n with Opus 4.7 as the brain. It self-verifies outputs so you don't need a

🚨 Google Gemini can now teach business strategies better than Harvard MBA professors, for free.

Copy these 7 prompts:￼

@RayFernando1337 Noticed that too, immediately had the same reaction as you. Very lame. Given how poorly the gpt 5 autorouter worked, I don’t have high hopes for anything where I can’t directly control at least the reasoning level.

Currently there's no access to Llama 3.1 405B, Meta's base model that came out in July 2024. I think it's a real shame (not alone either, cc @Sauers_ and @voooooogel, @repligate et al).

It's important that the public be able to experience what a large base model is like. In the

CLAUDE OPUS 4.7 IS OUT AND IT BURNS TOKENS LIKE NEVER BEFORE..

You used to send a prompt. Tokens were fine.

That's over..

New tokenizer eats up to 35% more tokens on the exact same text.

So Anthropic did something unusual - raised rate limits for ALL subscribers on the spot.

@nthnl_271 do not underestimate elon and china. the qwen model that came out today is better than opus 4.5 on benchmarks right now running on a 3090. If any of them slip there's plenty of people around to pick up the slack

5/7

Direct Comparison vs GPT-4o / GPT-5 series
• Opus 4.7 outperforms in complex multi-file coding and long-horizon agent tasks
• GPT models remain faster and more cost-effective for simple queries
• On hard professional knowledge work, Opus 4.7 requires noticeably fewer

> life after my brain gets connected to Claude Opus 4.7 by wearing a beanie

Haven't been frustrated with a model release since GPT-5, but Opus 4.7 has me there. I'm sure I'll figure out how to work with it, but right now it's doing a bunch of things that are difficult to taxonomize that are bugging me. Maybe it's overeagerness -- going beyond the ask.

Benchmarks suggest it beats Opus 4.6, GPT-5.4and Gemini 3.1 Pro

Agentic coding, tool use, computer use. Clean sweep on everything that maps to actual dev work.

@AnthropicAI coming for your job.

@mhmazur I focus heavily on AI Art so its cool to see this.  Claude is quite good with 3D worlds actually 

But Gemini doesnt surprise me with Google's access to imagery to train  Gemini on.

There is no way to still benchmark Diffusion Models

Thats something I want to build later this

Rather than investing years of effort and billions of dollars in frontier research, China-backed AI companies like DeepSeek are feeding outputs from leading American AI systems into their training pipeline, illicitly reverse-engineering the capabilities without including the same

@PrismML this is so nice 16x smaller size model with performance of qwen 3 

opensource is on another level 

this is the way ai should progress building better efficient models than just bloating it to increase the size and requiring high amount of GPU

Potentially the biggest day in AI so far this year

3 major releases you need to be trying immediately:

🔶 Claude Opus 4.7

• More agentic model. Front load tasks and information, then let it rip for hours

• Use auto mode to cut cycle time. Doesn't ask for permission

Introducing Claude Opus 4.7, our most capable Opus model yet.

Anthropic just did something unprecedented.

They released Claude Opus 4.7—beating GPT-5.4 on SWE-bench (64.3% vs 57.7%)—while admitting they're sitting on something far more powerful.

Claude Mythos is too dangerous to release.

Here's what that means 🧵

Claude Opus 4.7 is here!
Better coding on the hardest tasks, sharper vision (3.75MP images), and a new "xhigh" effort level. 
Same pricing as 4.6: $5/$25 per M tokens.

@safwan_sufiyan @grok That’s a good one too, google Gemini. Tried using to answer some quiz and it was cool.

look... perplexity drops "personal computer" ai for my mac files, qwen unleashes moE coding beast... but why's everything locked behind max subs or beastly GPUs? indie devs just wanna tinker without the paywall blues... sigh.

180 tok/s generation on a 4090 with qwen 3.6. if you're on a 4090 and not running this model yet you're leaving performance on the table. 3B active params at that speed is insane for agentic coding. thanks for the data @ErdalToprak, adding this to the comparison sheet

We upgraded Synter's AI agent to Claude Opus 4.7.

Ran head-to-head benchmarks against Sonnet 4.6 across 6 real ad strategy scenarios.

Here's what changed.

Claude Opus 4.7 is now live in Rork!

Smartest model from Anthropic, especially great in design

Claude Opus 4.7 is now live in Rork.

Anthropic's latest model with state-of-the-art coding and 3x sharper vision. Low-effort 4.7 matches medium-effort 4.6, so you can build more per session. Best model for design in our benchmarks

Day 0 access to Claude Opus 4.7 on AI/ML API!

Claude Opus 4.7 just dropped - and it's the model topping every benchmark out there.

@rvalyi @TiHIMothee @ThePrimeagen Despite what people say, I think Gemini 3.1 Pro is one of the best models in terms of nuanced understanding, and it's behavior was much more stable since the release compared to the other ones

Google's Gemini 3.1 Flash TTS just dropped: 70+ languages, natural-language vocal control, Elo 1,211 on TTS leaderboards, SynthID watermarking built-in. Text-to-speech just became production-grade for global products. #AI #voice

Opus 4.7 shipped today.

Same price as 4.6. Different model.

Here's what same price bought:

→ SWE-bench Pro: 64.3% - up from 53.4% on 4.6. Beats GPT-5.4 (57.7%) and Gemini 3.1 Pro (54.2%).
→ SWE-bench Verified: 87.6% - up from 80.8%.
→ CursorBench: 70% - up from 58%.
→

What makes this important is not just that Qwen dropped another model.

It is that a very serious capability is landing in the open.

That changes the emotional meaning of AI for ordinary builders.

For a long time, the best models felt like rented power. You pay every month. You

Open source AI is moving incredibly fast. Qwen 3.6 dropping with these benchmarks is a wake-up call for closed-model providers. Competition from the open community is only going to intensify.

Last 7 days trend