we compared four open-weight language models, gpt-oss-20b, Qwen 3.8 27B, Gemma 4 31B and Gemma 4 26B A4B, to find out which one fits which real-world job. Each model answered the same set of prompts across five categories: reasoning, factual accuracy, instruction following, coding, and Indian languages (including Hindi and Hinglish). Every category has five prompts worth 20 points each, so a model can score up to 100 per category and 500 overall.
To keep the test fair, every model got the same prompts and system prompt, each prompt ran in a fresh chat, and two people scored every response. We then turned the scores into a scorecard, a best/weakest/typical-failure profile for each model, and a recommendation for three use cases: a student help-desk assistant, a coding helper, and an Indian-language app. In our results, Gemma 4 31B came out strongest overall, with Qwen 3.8 27B as the runner-up for coding.