Patrick Daether

@thdxr I guess it depends on the models. With models like Kimi K3 in the ring, this could make sense in the near future.

i know this isn't the most appealing thing ever but i'm curious on the state of things would you pay for a $200/month sub that gives you access only to opensource models? say you got ~$800/month in usage

@SherryYanJiang I wasn't able to notice any meaningful difference between high and extra high or above. So I stay on high basically all the time.

does someone have a use case where they use gpt 5.6 sol extra high or ultra, and find that it's worth it??

@larsbuilds I dropped my Claude subscription for now. I'm all in for Codex + Kimi K3 just for Frontend.

I guess I have to reconsider my AI tooling until now I had > $20 ChatGPT sub > $20 Cursor sub (with up to $30 additional usage) > $10 opencode Go But with Fable staying in Claude Max I probably switch to this setup > Claude Max $100 > ChatGPT Plus $20 Still $40 more than my setup before but I am positive I can get that in with more productivity 😁

A news story in the German media. As a developer, this seems very outdated. I'd say Anthropic's reputation is much worse, and since GPT-5.6, the trend has been shifting from Anthropic -> OpenAI.

Today I cancelled my Claude Code subscription. Codex + Kimi works great for me atm. All this back-and-forth over access to Fable 5 is just annoying. ... and Fable 5 just isn't THE model anymore, compared to the others.

@PovilasKorop Yeah, it's crazy how fast things change these days.

@pdaether K3 wasn't released at the time of this video. Planning to re-test with K3.

I've being doing LLM benchmarks for FRONTIER models like Sol/Fable. Here's one screenshot from a video, from one of three projects for fixing OSS bugs. - Sol/Fable/Opus are similar, difference is PRICE - Grok 4.5 is VERY close to those - All other LLMs are VERY far from those

@GrahamJCampbell Interesting how different the experience can be. With GPT 5.6, you really need to set boundaries. Where Fable stops early, GPT often does too much. Besides that I really like to work with GTP 5.6

Really looking forward to Opus 5. GPT 5.6 is unusable for real work that is not just vibe coded slop with compaction every 69 seconds.

Wow, Kimi K3 is really good at frontend design. As we all know, tastes vary, but in my opinion, K3 is better than Fable 5 in terms of taste.

So if you compare it to GPT 5.6, I don't see to many scenarios in where to use it. Maybe for Frontend, as it seems very good in UI.

I tried out Kimi 3 with their basic tier. Very interesting model, bur I just reached my 5h limit quickly and also 22% of my weekly limit.

@Makuh90 If it is smarter maybe depends on the use case, but in my experience, it's definitely not faster.

But. Fable is far faster and smarter. And a much better writer.

@CommandCodeAI 👋 It's not official published yet. Is there a preview?

Who wants access to Kimi K3 model?

The trio of Sol, Terra, and Laun always provides just the right model for the job. Their strength lies in finding the right combination.

@BraedendotTECH That doesn't sound much different from the reviews you get from some human developers. 😀

GPT 5.6 SOL is funny You tell it to write some code You ask it to review that code It finds bugs in the code it just wrote… It fixes it… You ask again to review It finds more bugs… And it never ends 😂

@larsbuilds @tweetmarkdown Congratulations! 💪

a lot has happened on tweet.md since I launched it 2 months ago! beside multiple improvements and new features tweet.md already has 15 paying customers and over 4K conversions made so far 🚀 I truly appreciate anyone who bought a credit pack or helped me during the launch 🙌 I will continue improving 💪 thank you! 🫶

Perfect. 😒

@johncrickett I think the assumption that it reflects the average of all code is wrong. A good AI also learned to distinguish good and bad code, meaning the generated code is above average.

“AI writes better code than I do” is a strange thing to say when you think about it. AI was trained on a vast corpus of published code. It reflects the average of all that code. So what you’re really saying is: “My code is below average.”

@danclipping SEND

This video costs $0.4 not an actor. not a shoot. not even real people. gpt 5.6 + seedance 2.0 rebuilt them from scratch and I genuinely cannot tell the difference anymore most people are still paying $150/video for UGC creators while this system does it for pennies and scales infinitely I put together the exact prompting system that makes this possible not just prompts, the actual method so you can build your own characters and never run out of content comment "SEND" and I'll DM it to you IMMEDIATELY (most be following so I can DM)

@VadimStrizheus Too early to say as I need to test it a little bit more, but at the moment I would go with the GPT 5.6 family. Just compare the result from this DeepSWE v1.1 and have a look at the costs...

As a founder, which AI model do you prefer? 1. Fable 5 2. GPT 5.6 3. GLM 5.2