GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
Perciqa
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
Perciqa
Model Tree
Qwen/Qwen3-Coder-30B-A3B-Instruct
Input Modalities
Output Modalities
Supported Functionality
bash
pip install transformers>=4.51.0
python
from transformers import AutoModelForCausalLM, AutoTokenizermodel_name = "perciqa/Perciqa-Aurora-Code"tokenizer = AutoTokenizer.from_pretrained(model_name)model = AutoModelForCausalLM.from_pretrained(model_name,torch_dtype="auto",device_map="auto")system_prompt = "You are Aurora, an AI code assistant built by Perciqa. You help developers write, review, and understand code. You provide clear, correct, and complete solutions. When you're unsure, you say so."prompt = "Write a Python function to merge two sorted lists."messages = [{"role": "system", "content": system_prompt},{"role": "user", "content": prompt}]text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)inputs = tokenizer([text], return_tensors="pt").to(model.device)outputs = model.generate(**inputs, max_new_tokens=2048)print(tokenizer.decode(outputs[0][len(inputs.input_ids[0]):], skip_special_tokens=True))
bash
vllm serve perciqa/Perciqa-Aurora-Code
Then call via OpenAI-compatible API:
python
from openai import OpenAIclient = OpenAI(base_url="http://localhost:8000/v1", api_key="not-needed")response = client.chat.completions.create(model="perciqa/Perciqa-Aurora-Code",messages=[{"role": "system", "content": "You are Aurora, an AI code assistant built by Perciqa."},{"role": "user", "content": "Write a binary search in Rust."}])print(response.choices[0].message.content)
Monitor and trace every Aurora call with the Argus reliability engine:
python
import argusfrom openai import OpenAIargus.init(server_url="http://localhost:8000", agent_name="aurora-coder")client = OpenAI(base_url="http://localhost:11434/v1", api_key="ollama")@argus.trace(kind="agent")def aurora_code(query: str) -> str:response = client.chat.completions.create(model="perciqa/Perciqa-Aurora-Code",messages=[{"role": "user", "content": query}])return response.choices[0].message.contentaurora_code("Refactor this function to use async/await...")
| Benchmark | Score |
|---|---|
| LiveCodeBench v6 | 43.4% |
| SWE-bench Verified | 42.2% |
| HumanEval+ | 87.2% |
| MBPP+ | 80.6% |
| MultiPL-E (Python) | 72.6% |
| Field | Value |
|---|---|
| Architecture | Mixture-of-Experts (MoE) |
| Total Parameters | 30.5B |
| Activated Parameters | 3.3B |
| Layers | 48 |
| Attention Heads | 32 (Q) / 4 (KV), GQA |
| Number of Experts | 128 |
| Activated Experts | 8 |
| Context Length | 262,144 natively (up to 1M with YaRN) |
| License | Apache 2.0 |
Perciqa-Aurora-Code v1 is based on Qwen3-Coder-30B-A3B-Instruct by the Qwen Team (Alibaba Cloud).
This v1 release is a transparent rebrand -- the model weights are identical to the base model.
We've added:
@argus.trace)For the base model details: Qwen/Qwen3-Coder-30B-A3B-Instruct
| Version | What | Timeline |
|---|---|---|
| v1 | Transparent rebrand + Argus integration | Current |
| v2 | QLoRA fine-tune on synthetic coding data | Q3 2026 |
| v3 | Full fine-tune with expanded dataset | Q4 2026 |
Apache 2.0
bibtex
@misc{qwen3technicalreport,title={Qwen3 Technical Report},author={Qwen Team},year={2025},eprint={2505.09388},archivePrefix={arXiv},primaryClass={cs.CL},url={https://arxiv.org/abs/2505.09388},}