Milan

@LottoLabs @ThePrimeagen Comparable to say haiku 4.5?

@ThePrimeagen 27b is the best model release of the year so far

@0xSero Everyone's talking about q3.5 27b model, if you have used to enough, what you can compare it to, haiku 4.5?

Look at these and read them carefully, this will give you an idea of what you can expect from different Open Weight models. Qwen3.5-27B is the smartest Qwen3.5-35B-3A is the fastest huggingface.co/Qwen/Qwen3.5-2… huggingface.co/Qwen/Qwen3.5-3… Any Laptop, Macbook, or GPU should handle em

@dotta paperclip doesnt support hermes agent though right

Paperclip and Hermes Agents for Biotech Research

@LottoLabs Qwen 27b with Hermes agent, how good at a high level? Usable?

Qwen 3.5 27b and Hermes Agent ran through and a/b tested some architecture changes in my small tinygrad gpt model. I let it make the decisions fully and it ran to completion with minor steering. Haven’t looked at the code yet, I am skeptical but if it runs on CPU it’ll run anywhere. Obvious that larger params would have better loss/val. Interesting times.

@HarveenChadha Better name it, SharmajiKaLadkaGPT

we tested Sarvam 105B on the JEE Mains 2026 paper conducted on 28 January 2026 it scored 70/75 on pass@1 and 75/75 on pass@2

@HarveenChadha Way low, the memory bandwidth

Should I ??

@ivanfioravanti Any eta on m5 ultra studio?

I hope M5 Ultra will be good enough to meet the hype around it 👀

@HarveenChadha Congratulations sardarji, very happy for you and sarvam.

10 months back parents were not happy when I left MS Today when I reached, they were smiling Dad showed me all the news channel recordings, newspapers mentions of sarvam Mom told me how she promoted sarvam in whatsapp groups and to neighbours Overall, a very small win but a long way to go

@DorianDevelops Kimi 2.5

MiniMax M2.5 or Kimi K2.5?

@ashen_one 512 M5 ultra would be able to run minimax and glm5 quant, so one should go and buy 512 memory for sure

Should you really buy a Mac Studio unless it's the maxed-out 512 GB version? I have been talking to QWEN, a local LLM on this $4,000 98gb Mac Studio for only 2 hours, and I hate him. He's so stupid. He's actually so slow and so dumb that my main open claw, that's running Opus 4.6, yells at him in front of me Like, even to get a Discord message, it's taking a minute plus Unless you have 512GB of RAM on that big fat Mac Studio, I don't think running local LLMs is a good idea. I think it's worth it to just eat the cost of any API that you're using and be able to switch to the newest APIs like Minimax or KIMIK, and not limit yourself to wasting hardware Even though I have this 98 GB Mac Studio that costs $4,000, the local LLMs that are allowed to run on it are so dumb and slow that it makes more sense to just pay $100/month to access whatever Chinese model is better through their API If you really want to experiment and learn and have fun, you should only really do that with the 512 GB Mac Studio so that you can run actually good and smart models compared to these slow and stupid ones it be like that. I'm still having fun, and it was fun setting it up But this is just me publicly learning less I usually do on this account

@Dating_Dynamics Its a simple fact, coz of olfactory adaptation

A sex therapist with 30 years experience revealed the real reasons so many couples go from passionate to platonic in under 2 years. It's not stress, kids, or age. It's these seven dynamics shifts…

@ivanfioravanti @Apple Any date we have as yet, even credible rumors?

As soon as M5 Max and Ultra will be released I’ll buy 1 + 2. So please @Apple release them before end of March to boost your Q1 earnings 😎

@applesclubs Memory bandwidth?

Apple’s next powerhouse is loading… 💻⚡️ New Mac Studio could arrive after early March 2026 with: • M5 Max & M5 Ultra options • Faster SSD speeds • Same compact “squircle” design • Starting around $1,999 No redesign — just pure performance upgrades. Are you waiting for M5 Ultra? 👀🔥

@stolinski Most likely in june

This Mac Studio M5 Ultra needs to drop soon.

@simicvm @Prince_Canuma Compared this asr with whisper, parakeet etc?

fully local Wispr alternative app running Qwen3 ASR 0.6B via mlx-audio-swift. practically instantaneous result in any active input field. @Prince_Canuma

@test_tm7873 Benchmaxxed obviously

?!?! Wtf. That is Strong! Small model. Or benchmaxxed.

@TheAhmadOsman How much memory needed to run this at say q4 with full context,

we have opensource Opus 4.5 at home now Zhipu AI cooked with GLM-5

@TheAhmadOsman Rtx 6000 pro vs 2x dgx spark, what would you suggest, equal price, run smaller models faster vs big models slow?

Prediction We will have Claude Code + Opus 4.5 quality (not nerfed) models running locally at home on a single RTX PRO 6000 before the end of the year

@iotcoi In our experience how good really is the model? Close to sonnet 4.5?

I just woke up Claude Code Agent Swarm on local Qwen3 Coder Next. No cloud. No Internet. No quota anxiety. No 'You've hit your limit, resets 10 pm' One GB10 GPU - 100 tokens/sec generation - 17,871 tokens/sec read top speed - 256k context window - Swarm tool calling just works

@andimarafioti Would be great to do a comparison of available open souces asr now, we got few good ones, qwen, glm, parakeet to name few

This is getting crazy good. Running locally on my laptop with streaming transcriptions without compromising quality.