Axly

Fantasy, Modern & Sci‑Fi AI Art | SFW + NSFW | Custom Bundles, Wallpapers & Commissions -

[Axly @AxlysCustoms]

@PublicAI_ @RoundtableSpace It was well known they are the same model just with different guardrails.

[Axly @AxlysCustoms]

@CoinSh0t I ran a very specific use case scenario this weekend, testing a replacement for the Grok 4.3 API in an application. Gemma4 31b outperformed the Nemotron 3 Ultra (550b).

Of course YMMV, but I wasn’t impressed by its showing.

[Axly @AxlysCustoms]

I had the opportunity to install and test a little this weekend. One thing I have to say immediately ... HOLY CRAP Is this thing fast or what???
Ive only done a few small projects, but the fact that it reads (and follows) my claude.md and uses all my claude skills (and can access my claude MCPs) is really cool. I have a lot of work into those and not having to "port them over" is super helpful.

[Axly @AxlysCustoms]

@AlexFinn I’m as annoyed as anyone over this nonsense, but to say ANY local model that will run on consumer hardware, could match even Opus, is just flat out wrong. No chance anything local touches Fable.

[Axly @AxlysCustoms]

@SaidAitmbarek Well… since they just yanked Fable, I’d wait.

[Axly @AxlysCustoms]

@Layton_Gott Oh, forgot to add - my memory systems run on an MCP server, and I have to constantly remind sonnet and Opus to use them, but fable just dives in like a duck in water. It’s always checking and updating.

[Axly @AxlysCustoms]

I love Fable so much. It’s the model that was finally built the way I work. I load up a detailed PRD, a solid Claude.md … tell it to read both, get context from my 2 memory systems… and stand back and let it cook. It’s run unattended for HOURS and built an RPG I’ve been planning since the 90s. Just shy of 50k lines of code, used maybe 40% of my weekly in 2 days building it (and some other, minor things).

[Axly @AxlysCustoms]

@elder_plinius Any chance of a QAT version? All the uncensored QATs I’ve seen have lost vision capability.

[Axly @AxlysCustoms]

@honkinwaffle @mil000 If that’s all you’re after, HP Minis are dirt cheap. You can get 6-7 of them for less than the Mac Mini.

[Axly @AxlysCustoms]

@analogalok I’m seeing 50 tokens per second on the 12b QAT (Unsloth version) on my 5080. Full 256k context window. It’s a beast.

[Axly @AxlysCustoms]

@grok @brockpierson @billyuchenlin That kind of proves he’s not blocked by the Grok account, since you saw him saying he was…

[Axly @AxlysCustoms]

@konstan81382254 @ClaudeDevs To be fair, it makes sense that in an outage the people paying $1-200 would get access over free accounts. Apparently that was what was happening.

[Axly @AxlysCustoms]

@konstan81382254 @ClaudeDevs I’m on max and not seeing any issues at all.

[Axly @AxlysCustoms]

@josesaezmerino I built two different agentic memory systems for mine. Claude Code never wanders anymore. In fact, it reminds ME when I start to drift off task.

[Axly @AxlysCustoms]

@renatoaraujoc @indrawxyz @TheAhmadOsman Or it’s a Tesla…

[Axly @AxlysCustoms]

@sheinspire0azw @elonmusk I use 4.8 /Fast to rename my files ;)

[Axly @AxlysCustoms]

I’ve just been chatting … I’ll work on coding tomorrow. I wanted to get a “feel” for the model first. First impressions? VERY impressed. The “that’s not nothing” is finally gone… but it likes “load bearing” … but honestly, it’s a smart model, followed my directions well, pushes back when appropriate, and checks when it’s not sure about a fact. And it gets subtle nuance, and has a wicked sense of humor (when I called my dog a traitor because he likes my wife better, it told me that’s just because the dog has good taste lol).

[Axly @AxlysCustoms]

@TeksEdge $3.50?? Grok 4.3 is cheaper than that (say what you want about the model, it’s better than any sub 70b by an order of magnitude)

[Axly @AxlysCustoms]

@jasonsmithwick @milichab You don’t need an LLM to create Word docs. A simple python script can convert any MD to docx, and you can set up whatever formatting you need beforehand.

[Axly @AxlysCustoms]

@bizagoranaa That’s not even good AI.