GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
Miiyamoto255
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
Miiyamoto255
Model Tree
Input Modalities
Output Modalities
Supported Functionality
Aetheris-Core-2B is a compact, from-scratch 2-billion-parameter language model based on the Gemma2ForCausalLM architecture. It is designed for fast, capable, and flexible local inference. It is built with a strong emphasis on coding, reasoning, creative generation, instruction following, and open-ended problem solving.
Unlike heavily constrained assistant-style models, Aetheris-Core-2B is designed to minimize unnecessary refusals and avoid blindly rejecting legitimate technical, creative, educational, and research-oriented requests. The goal is to make the model feel direct, useful, and developer-friendly while still maintaining sensible boundaries around genuinely harmful requests.
💻 Programming & software engineering: Python, JavaScript, HTML/CSS, C/C++, debugging, algorithms, game development, and code architecture.
🧠 Reasoning: Multi-step analysis, problem decomposition, troubleshooting, and technical explanations.
🔧 Developer assistance: Refactoring, optimization, error diagnosis, project planning, and implementation guidance.
🎮 Game development: Basic-medium gameplay systems, mechanics, UI, level design, scripting, and asset concepts.
✍️ Creative generation: Characters, worlds, dialogue, stories, concepts, and brainstorming.
📚 General knowledge: Explanations, summaries, learning assistance, and question answering.
⚡ Local inference: The 2B parameter size is intended to provide a practical balance between capability, memory usage, and inference speed.
Aetheris-Core-2B aims to be useful before being restrictive. It should interpret requests in context rather than treating individual keywords as automatic reasons for refusal. Legitimate experimentation, reverse engineering for learning, programming research, fictional scenarios, and unconventional ideas should be handled intelligently instead of being unnecessarily blocked.
The model is also designed to communicate clearly and efficiently. It should avoid excessive disclaimers, repetitive warnings, artificial enthusiasm, and unnecessary verbosity when a straightforward technical answer is sufficient.
This model is licensed in MIT License. You can use, modify, copy, quantize this model, but you should add Miiyamoto255 in credits