Aspiring MLE and interested in Data Science
MilkyWay
Joined June 2022
- Tweets1.5K
- Following137
- Followers34
- Likes1.5K
In the near future, I hope small models will be so good enough that they can be run locally on capable smartphone devices
Today, we’re releasing Bonsai 2 2B, 4B, and 9B.
Based on Qwen3.5, these new ternary models bring stronger reasoning, visual understanding, and tool calling to compact footprints designed for everyday devices:
• 2B: 0.43GB
• 4B: 0.95GB
• 9B: 2.0GB
For this release, we focused on what makes small models useful as local assistants: following instructions, avoiding repetitive loops, calling tools on mobile devices, and giving useful answers without unnecessary output. The 9B model uses roughly 40% fewer tokens on standard instruction-following and everyday math tasks. Solving the Riemann hypothesis can wait.
These models are successors to our first-generation Bonsai 1.7B, 4B, and 8B, and provide a foundation for bringing practical agentic functionality to local applications on personal computers, phones, and wearables.
The Bonsai 2 family now includes four sizes spanning 2B to 27B parameters. Developers can choose the balance of capability, memory use, and responsiveness that fits their application and hardware.
Available today under Apache 2.0.
Karak S retweeted
A 35B language model running on an iPhone using only 1–2.5 GB of peak memory.
No cloud. No remote server. No desktop GPU.
Today, we’re open-sourcing Edge0 — a framework for running large AI models fully on-device.
Karak S retweeted
GLM-5.3-Flash. Uncensored. Native FP8. 🐳
We just released OrcaRouter’s uncensored weights for GLM-5.3-Flash — 320B parameters / 18B active, directly at the original block-FP8 precision.
No LoRA. No jailbreak prompt. Refusal removal is baked directly into the weights.
The evals are particularly interesting:
→ MaliciousInstruct refusal: 96% → 11%
→ JailbreakBench: 93% → 12%
→ AdvBench: 97% → 15%
→ HarmBench: 93% → 18%
→ XSTest benign over-refusal: 2.4% → 0.4%
But refusal does not go uniformly to zero.
Our experiments suggest part of GLM-5.3-Flash's alignment is not mediated by a single linear refusal direction — meaning Z.ai may have built a substantially deeper refusal mechanism than we usually see.
That makes this release interesting beyond uncensoring: it's a useful artifact for studying how frontier-model alignment is actually represented inside the network.
Released for AI safety, interpretability, red/blue-team and refusal-mechanism research.
Weights on Hugging Face: huggingface.co/orcarouter/GL…
API (official weight): orcarouter.ai/models/z-ai/gl…
GGUF, MLX and other quantized formats coming soon.
Karak S retweeted
Telegram has applied for the .gram domain zone.
If the application is approved by ICANN, a billion Telegram users could get their own second-level domains — yourname.gram.
Users would be able to set up their interactive websites hosted by Telegram — with one prompt ✨
And it's Opensource
Exciting news: GLM-5.2 (Max) ranks #2 in Code Arena: Frontend, with +29pt over Claude Opus 4.7 (Thinking) and only behind Fable 5! GLM-5.2 is the best open model vs Kimi-K2.6 and Minimax-M3 by a large margin.
- #2 React and #4 HTML sub-leaderboards
- Ranks as the top model in nearly all sub categories: Brand & Marketing, Reference-Based Design, Data & Analytics, Consumer Product, Gaming, and Simulations.
Congrats @Zai_org for the incredible milestone!
Opensource 👌
BREAKING: GLM-5.2 is now 1st on Design Arena.
With an Elo of 1360, GLM-5.2 has jumped ahead of the now unavailable Claude Fable 5.
And it's open weights.
This is an improvement of 4 positions and 27 Elo points to achieve one of the highest Elo scores in our code categories since Design Arena started.
Huge congratulations to the @Zai_org on the release!
Great +_+
Tired of your token costs?
M3 is free on @PBDTokenRouter through june 17th
With every other MiniMax model 50% off 🔥
That means building with 1M context, and $0 in / $0 out
Build something today, before this opportunity ends.
Nice 👀
🚀 MiMo Code V0.1 is now live and open-source！
More than an AI coding assistant in your terminal — it's the smartest coding partner you'll ever work with.
Comes with MiMo V2.5, a multimodal model available free for a limited time, featuring a million-token context window—ready to use out of the box.
♾️ Infinite Context: Knowledge accumulates automatically, and with lossless compression, even million-line projects keep every critical detail intact—quality never drops.
🧠 Agent-Model Synergy: An Agent framework deeply optimized for MiMo, with a full closed loop of testing, review, and validation—so complex tasks get done in one pass.
📝 Compose Mode: Specs → Plans → Build → Report. Design first, code second—clear thinking, no rework.
🔄 Self-Evolving System: Every session is automatically reviewed, distilling experience and best practices—the more you use it, the smarter it gets.
🎙️ Voice Input: Powered by MiMo-V2.5-ASR — just speak instead of type, and your voice becomes the prompt for truly hands-free coding.
🔌 Claude Code Compatible: Automatically loads your existing skills, MCP servers and commands, and reuses your API configuration—zero-cost migration, no setup required.
🌐 Open & Flexible: MIT licensed, with support for leading model providers including Anthropic, OpenAI, DeepSeek, Kimi, GLM and more.
Install in one line:
Mac & Linux
curl -fsSL mimo.xiaomi.com/install | bash
(For the best experience，we recommand Mac user use it on iTerm or vscode terminal)
Windows
npm install -g @mimo-ai/cli
🔗 Learn more
Website ↓
mimo.xiaomi.com/mimocode
Blog ↓
mimo.xiaomi.com/zh/blog/mimo…
GitHub ↓
github.com/XiaomiMiMo/MiMo-C…
I see most of my peers are actually distracted into different stuffs, let's see what happens in the end
If you're vibe coding apps or websites and don't know manual debugging or haven't any idea of core of any programming language properly,it'll be useless for you then!
Karak S retweeted
While making an exit, collect your documents lads.
Pay Slips. Employment contract. Appraisal Letters. Reliving letter. Letter of Recommendation (if you can). A LinkedIn recommendation (if you can).
And most importantly, think thrice before you think about leaving on a wrong note. The world is very small lads, and you never know who you'll run into again. Maintain a good, clean, professional relationship with your ex-colleagues.
Actually keep in touch with your ex-teammates. Who knows, you'll get laid off at your new org and you ae looking for jobs again! A bleak bridge is better than a burnt bridge after all, and these things end up going a long, long way.
