Model Guide: Pros & Cons : campusGenAI : UMass Amherst

Compare different models based on their relative strengths and weaknesses.

Model	Pros	Cons
GPT-5	Suitable for a wide range of tasks, including higher level tasks. Ideal for PDF and/or image comprehension. Offers complex reasoning and in-depth analysis. Supports coding workflows.	Other models by be more suitable for more specialized tasks. See Model Quick Guide for more in-depth descriptions of the models and their specialities. Uses more credits on average.
GPT-5-mini	Higher speed and efficiency than other GPT models. Ideal for users that are new to using GenAI. Designed for smaller tasks. Uses less credits on average.	Not the best for more high-level tasks. Other models may be more suitable for more specialized tasks.
GPT-5 Codex	Specialized for software engineering and coding tasks. Supports interactive coding sessions, including assistance, debugging, and refactoring. Also capable of independent code generation.	Not the best for non-coding based tasks. Slightly higher latency on average.
DeepSeek-R1	Designed for high-level and complex tasks. Uses deep reasoning in answer generation.	Higher latency on average due to the model's deep thinking. Uses more credits on average.
Claude 4.5 Sonnet	Specialized for coding, reasoning, and mathematics. Provides advanced capabilities in interacting and using computers. Ideal base model for a complex agent. Designed for large- to moderately-sized tasks.	May need user intervention to avoid "overwriting" (generating unnecessary documentation and reports on the given task). Slightly higher latency on average.
Claude 4.5 Haiku	Ideal for smaller tasks, such as interactive coding sessions. Provides advanced capabilities in interacting and using computers. Optimized for speed and efficiency. Uses less credits on average.	Not the best model for large-scale tasks.
LLaMA 3 70B Instruct	Lower latency on average. Prioritizes speed and efficiency. Lighter-weight model designed for smaller tasks.	Responses can sometimes be long and unclear.
Web Search AI	Can use and report on information from the Internet. Ideal for prompts needing more up-to-date information such as news and current events.	Higher latency on average. Possible pulling from unreliable sources on the Internet.
Mistral Small 2402	Cost effective (uses less credits on average). Designed for smaller tasks.	Not the best for high-level tasks.
Grok 3	Integrated with X (formerly known as Twitter). Knowledge of current events and pop culture.	Not the best for tasks that require complex reasoning.
DALL-E 3 Image Generator	Highly specialized for image generation.	Resultant images may have qualities indicative of AI usage. Uses more credits on average.