--- language: - ar - en - fr license: mit library_name: transformers tags: - foundational-model - coding-agent - software-engineering - universal-tool-calling - 256k-context - swe-bench - genisi-codex-open - anesnt - algeria - batna --- # 👑 Genisi Codex Open (9B) — Foundational Autonomous Software Engineering Model **Genisi Codex Open (9B)** is a foundational, open-weight, multilingual (Arabic, English, French) autonomous software engineering AI model and full-stack code generator trained and developed from the ground up by **AnesNT**, an Algerian AI research project from **Batna Province, Algeria (ولاية باتنة - الجزائر)** founded by **Anes Kameche (أنس كامش)** at age 15. ![Genisi Codex Open Benchmarks](./Genisi_Codex_Open_Benchmarks.png) --- ## 🇩🇿 About AnesNT & The Creator (نبذة عن المشروع والمؤسس) - **Model:** **Genisi Codex Open (9B)** (Foundational Autonomous Coding Model) - **Research Project:** **AnesNT** (مشروع ذكاء اصطناعي جزائري مستقل) - **Founder & Lead Architect:** **Anes Kameche (أنس كامش)** — 15-year-old Algerian AI researcher & software engineer. - **Location:** **Batna Province, Algeria (ولاية باتنة، الجزائر) 🇩🇿** --- ## ⚡ Foundational Specifications - **Parameters:** 9.46 Billion Dense Parameters - **Context Window:** **262,144 Tokens (256K Native Context)** - **Max Output Generation:** **32,768 Tokens (32K Output)** - **Native Reasoning:** Built-in `...` deep architectural planning and multi-turn verification. ## 📊 Global Benchmark Evaluation | Benchmark | **Genisi Codex Open (9B)** | Qwen3.5-9B | Gemma-4-31B | | :--- | :---: | :---: | :---: | | **SWE-bench Verified** | **72.1%** | 53.2% | 52.0% | | **SWE-bench Pro** | **48.8%** | 31.3% | 35.7% | | **NL2Repo** | **35.1%** | 16.2% | 15.5% | | **MCP-Atlas (Agent Tools)** | **58.4%** | 46.8% | 55.0% | | **GPQA Diamond** | **86.4%** | 81.7% | 84.3% | | **Terminal-Bench 2.1** | **47.9%** | 21.3% | 42.1% | ## 🚀 Quick Usage ```python from transformers import AutoModelForCausalLM, AutoTokenizer model_id = "AnesKAM/Genisi-Codex-Open" tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True) model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", trust_remote_code=True) ```