Deep 🔫

GPT-3.5 Turbo had a 16k context window, and at the time it felt like more than enough. A few years later, we're measuring context windows in the hundreds of thousands and even millions of tokens. AI progress has been moving fast.

@asheshtwt If keeping bench scores aside irl scenarios kimi k2.6 outperformed qwen 3.7max and also its 3x cheaper than qwen

@thephatcoder Why tho?

Past week i made some good software with Kimi 2.6 and loved it soo much ❤️ This week lets try Qwen 3.7 max 👀

@Harish_521 Damn, how do you guys get such domains how much does this costs 🫪

Claude FABLE 5 is insane. i know literally NOTHING about coding. ZERO. And i built this app in 1s http://localhost:3000/ check it out.

@rxhit05 Fableous

So has anyone tried the new claude fable? How is it

@dhruvtwt_ ATP Dario must have manually hardcoded the response of this brainrot question 😆

damn... fable 5 is scary good

@YashHustle_22 Kimi k2.6, as it accepts image as input which is quite helpful sometimes

What's your go-to open source AI model? - Qwen 3.5 - GLM-5 - Kimi k2.6 - DeepSeek V4

@OjasSharma276 I haven't tried 4.7 and 4.8 yet as well 🙃

Am I the only one who hasn't tried out Claude Fable 5 model yet?

Big pickle from opencode is really good model if your end goal is clear

@Snixtp Naah bro I don't believe that

Mythos this, Fable that Watch it get mogged by Qwen3.7 27B

@sflorimm Imo cursor and deepseek dont deserves the place in this list

Suddenly it hit me. What happened to DeepSeek? Sora? GitHub Copilot? Llama? Cursor? Perplexity? What happened?

@AlemTuzlak And that tweet wasn't that deep to deserve 10000 words of articles🫪

Its funny how some big AI influencer writes a random tweet and then everyone writes articles on what he meant exactly and how you can use XYZ to achieve what they said.

My current timeline is all about loop, and i am already bored with this tl can someone please change the topic

@adrianpkstream @darekgusto Yes totally agreed

@darekgusto @thephatcoder +1 to what Darek said. With a good reference implementation, strong linting, and exhaustive style and architecture guidelines, the code produced is often more uniform and consistent than the code I would have written by hand myself.

Code review fatigue is hitting hard these days. AI code is getting worse to review not better. Shipping 5+ features a day has turned reviews into a nightmare. We're drowning in PRs. Anyone else feeling this? How do you keep quality high?

@Tech_girlll I see vibe coding as giving the AI the destination, not the roadmap. You describe what you want, and the AI figures out the architecture, implementation, and decisions. Once you're heavily planning and steering those decisions, it's no longer pure vibe coding.

No one seems to agree on what vibe coding actually means. Is it coding without understanding? Coding with AI? Or just a catchy label for modern software development?

@Venkydotdev Prompting is an art 😆

Is prompting an engineering?

@mikeydsoftware Not necessarily faster, but with smaller loops and more focused goals. That way, if there's a loophole or the agent starts drifting, we can catch it early and correct it before it compounds into a bigger issue.

@thephatcoder you mean optimizing the agent loops to be faster?

Agent loop or Human in the loop? Which one produces better software today? 🤔

@Aevmorfop U mentioned exactly what I'm facing right now Sometimes I move so fast that I lose track of what's actually going on and it becomes hard to catch up with my own code and decisions I'll try following your advice shipping one feature at a time but with the highest quality possible

@thephatcoder Sometimes I purposefully go slower so I’m able to follow what’s going on. You don’t need 5 features a day. You need one that’s working super well.

@darekgusto True 🫶

@thephatcoder Everyone loves getting hearts

This X notification is really addictive