Introduction: Our project compares five open-weight AI models to find the best model for different tasks.

Problem: No single AI model is best at everything. We must compare accuracy, speed, cost, and hardware needs.

Models Selected: We evaluate OpenAI gpt-oss-20b, gpt-oss-120b, Qwen3.8-27B, Gemma 4 31B, and Gemma 4 26B A4B.

Testing Method: We use consistent prompts and settings to compare the models fairly.

Evaluation Criteria: We check reasoning, coding, instruction following, factual accuracy, JSON output, and efficiency.

Published Benchmarks: The presentation reports strong coding and science benchmark scores for Qwen3.8-27B. These are published results, not our own test results.

Testing Examples: We use maths problems, factual accuracy questions, JSON formatting tasks, and Python coding challenges.

Model Recommendations: gpt-oss-120b is a candidate for complex reasoning, Qwen3.8-27B for coding and science, and Gemma 4 31B for vision tasks.

Limitations: Actual team testing is still pending. Performance, speed, hardware requirements, and safety need further evaluation.

Conclusion: The best AI model depends on the task. We should choose models based on evidence and testing, not hype.