LobsterDAO 🦞

29 July 2026 09:50

hook chat-GPT web interface into github, have it review and plan using the web-based Pro model, then use codex to implement, commit, etc

Claude/Kimi/whatever can review code/PRs, call out discrepancies if needed (two 'brains' being better than one), but you can save massive amounts of Codex tokens by splitting work between the GPT web interface and the desktop/Codex app.

LobsterDAO 🦞

29 July 2026 02:51

you can also have fable delegate to opus 5, since they are almost two usage lanes

LobsterDAO 🦞

29 July 2026 02:01

what tasks? what context size etc? for me its still much more efficient to use fable than the other models. you got any tooling around your usage, or just rawdogging it?

LobsterDAO 🦞

29 July 2026 01:47

havent gone back to fable yet as it nuked my token limit in like 30 min lmao

LobsterDAO 🦞

29 July 2026 01:41

OPUS 5 for planning and implementation, Codex as the Software Engineer is how i set it up.

LobsterDAO 🦞

29 July 2026 01:40

Hey shit how they changed the subscription tier to access it tho

LobsterDAO 🦞

29 July 2026 01:06

Am i the only person who uses codex and finds it very good bang for my buck?

LobsterDAO 🦞

29 July 2026 00:57

It sucks compared to the other models, in the sense that it writes very shit code but everything is so much faster that it doesn’t matter

LobsterDAO 🦞

29 July 2026 00:33

Anthropic lately changed tokenizer and I also feel this way. Been playing with Grok 4.5 and Qwen 3.8 Max as Opus/Fable replacements

LobsterDAO 🦞

28 July 2026 21:49

But I’m physically in Europe but I passed KYC on my UA passport can bought

LobsterDAO 🦞

28 July 2026 21:48

I’m not eu resident so I can’t

LobsterDAO 🦞

28 July 2026 21:47

Can you try it rn with a small amt?

LobsterDAO 🦞

28 July 2026 21:47

I mean to buy stables with debit cards

LobsterDAO 🦞

28 July 2026 21:27

and anyone else not being able to onramp any cash via cards in EU rn?

LobsterDAO 🦞

28 July 2026 21:25

anyone see someone actually using te Privy Account linked card thingy?

LobsterDAO 🦞

29 July 2026 04:07

I use sonnet 5 xhigh and have to call fable 5 and opus 5 subagents to see where they disagree or catch different things. Then sonnet synthesizes.
I'm not convinced any of these models are that much different from each other besides costs. If you do most of the mental work they can all handle grunt stuff. If you don't want to do the mental work, well that's just kicking the can cause you'll end up having big slop to review otherwise.

LobsterDAO 🦞

29 July 2026 02:20

I use fable 5 at high or medium reasoning effort for planning/reviews/specs and delegate tasks to gpt 5.6 sol

LobsterDAO 🦞

29 July 2026 01:54

idk how any of yall are using Fable, we burn through tokens like zap

LobsterDAO 🦞

29 July 2026 01:47

fable for doing stuff overnight imo

LobsterDAO 🦞

29 July 2026 01:41

Codex certainly the best on the code side imho

LobsterDAO 🦞

29 July 2026 01:13

Codex is the best quality atm, grok is nice for speed/cost/quality balance from what I have heard.

LobsterDAO 🦞

29 July 2026 01:00

kimi v3 is super fast but then seems to get super slow if you use it a lot

LobsterDAO 🦞

29 July 2026 00:56

Yeah I love Grok, especially for that speed.

LobsterDAO 🦞

28 July 2026 23:09

Same @ivangbi

LobsterDAO 🦞

28 July 2026 21:49

Well, it works for me because I’m not eu resident

LobsterDAO 🦞

28 July 2026 21:48

What does it say to you? Whats an error

LobsterDAO 🦞

28 July 2026 21:47

Moon pay doesn’t work for you?

LobsterDAO 🦞

28 July 2026 21:40

You mean sell crypto and top up cards with fiat?

LobsterDAO 🦞

28 July 2026 21:27

Pics will be updated closer to Devcon + huge ecosystem map (3 by 2 meters).

For now -> list within the fileverse ddocs + git later

LobsterDAO 🦞

28 July 2026 21:20

Okay, a bit off topic but anyone else feels like they have started skimping on the Max/Pro subscriptions for both ChatGPT and Claude? I can't tell if this is because the newer models are eating through tokens in their thinking blocks or because they have actually reduced the limit.