Tech2Wild

Claude Opus 5 Just Dropped… And It Beat Fable 5?! youtu.be/7dHwVGf6P0E

@loktar00 yes

@Tech2Wild That is impressive as heck for having it at home. 4 sparks right?

Wanted To Try This Out on GLM 5.2 Local. One Shot in 5 Minutes... I'm Impressed....

Bruh Imagine One Of These Chasing You. Some REAL Black Mirror ISH

@u1tra_instinct This clip is wild

I need the reality bender installed to my skull…..

Bruh, I was going over a video with my Agent running Qwen 3.6 35B A3 Uncensored and was asking for feedback and send him a comment someone said and he just went off lol...

@rwtenjoyer That’s what I believe that they basically did not need any more money that they’re good with what they got

@Tech2Wild That they don't need any more funds right now + they rather not dilute themselves at current valuation (vs where they believe it will go, up).

What This Mean ?

@pardoxAI vllm

@Tech2Wild What engine ?

2x RTX 3090s running 32 coding agents in parallel. Task: Build a trading bot (handle partial fills, rate limits, race conditions + explain reasoning). 🚀 qwen3.6-35b-a3b-autoround on 3090 35B • Total: 76,183 tokens in 64s • Aggregate: 3051.1 tok/s peak · 1183.8 tok/s sustained • Per-stream: 101.2 high / 80.5 low / 96.8 avg tok/s • TTFT 0.25s avg · E2E 55s avg

The thing is @ArtificialAnlys did test it. That is why they put the cost per task at a lower rate for a reason. But let's agree to disagree. I gave you what a team of people who have been testing it for a week has to say and then you asked me if I tested it and I told you based off my observations I agree it is cheaper too. Then Anthropic said it as well so 3 sources so far say it's cheaper against your 1 source. At the end of the day its opinions but I will go where I see more sources and personal experiences are claiming the information. But your locked on your opinions like I am mine so no reason to go back and forth and rather we just disagree and move on. If your tiresome you might got to mute the accounts that make you feel that way.

The trouble with X and elsewhere is that this is the process: 1. New model drops; 2. Everyone raves at the claims; 3. No one actually tests it; 4. New model drops. Repeat from #1. It’s getting very tiresome to read a hundred people parrot the same claims for the next new model day in and day out. It’s starting to feel like "all hype and no substance". Heaven knows what newbies to AI and LLM think when they read all of these posts. It’s like we’re purposefully putting up barriers to exclude people. No one cares what a company claims. People only care about what the model does and what it actually costs in day-to-day usage. Your “1 day test opinion” tells you the cost is lower. My 1 day test opinion tells me it’s just as bad as Fable 5 - oddly, it seemed better last night but, this morning, it was just ran through its 5hr window in no time at all. In other words, anecdotal claims are not a useful metric.

Fable 5 was too powerful to just hand out. Limited access, usage bars, the whole rollout. Then Opus 5 drops, beats it on the benchmarks, and costs less. So which is it? Either I'm missing something or somebody's bench maxing. Explain it to me 👇

@ayayalar I am not doing 20 tok/s though

@Tech2Wild Utilizing 4 dgx spark to run a single model at 20 tok/sec tough pill to swallow.

So I’ve Worked with GLM 5.2 For 2 weeks now over Deepseek v4 and I just got to say this is a COMPLETE model. Great Agentic work, when it hits road blocks it doesn’t loop it finds the solution, or it admits it doesn’t know how to do it and allows me to escalate to Frontier.

Since Moving The 3090 In The Garage. There is a SLIGHT temp increase. Going to have to get a fan to blow to the side but for the most part it’s doing its job.

I am about to run this model at a nice currency of up to 4 with out a real drop in tok/s. I’ve allow my wife and friend of mine to have their agents connected to the model as well and all 3 of us have been using it without issues.

@TechMDAI Make sure the DCP and K5 is all matching me bro you can get that 30+

First 4 DGX spark boot. Working to get it to the target what @Tech2Wild has. But I am happy it working now. It’s very useable currently and I know it has more in the tank.

@NsitnovSitnov Your ready I see

Will 4 x DGX Sparks Be TOO Little ? Damn

@jonathan_conn @NsitnovSitnov @MarkSunner No but I do see a spot for my 4 Sparks. @MarkSunner we can run 12 together and sell GLM 5.2 out to people maybe even Kimi K3 🤣

@NsitnovSitnov @Tech2Wild @MarkSunner you guys roommates or something?

@codeRunnerUK I mean the model just dropped so we can only go based on what is said. As far as my one day with the model I have noticed a drop in usage overall. So if you want my 1 day testing opinion, yes it’s lower.

@Tech2Wild Ah, so no actually testing for yourself? Is everyone doing the same when they post on here? It’s starting to feel like I’m the only one out of the thousands of AI users on X that has this “token squeeze” issue.

Hmm So is Opus the NEW Sonnet and Sonnet is the new Haiku and Fable the NEW Opus ? lol...We will see the progression of each model. Maybe Fable 5.1 makes some crazy leaps that separates itself from Opus.

@codeRunnerUK We're getting it from them.... x.com/claudeai/statu…

What you’re missing is that everyone keeps saying it “costs less”. It doesn’t "cost less” when it hits its 5hr window in under an hour of doing fairly menial tasks on an existing code base. It’s doing just what Fable 5 did. Where is everyone getting this “costs less” from? Are you all on $1000/week plans or something?