Choose the right model for your task. A practical guide to the latest models in UCSB AI Commons.  

Understanding the model families

Expand each of the model families below to learn more.

Micro, Lite, and Pro

Think of the Nova family as your primary toolkit for general tasks.

Amazon Nova Micro: The fastest, most token-efficient model. Best for simple, high-volume tasks.

Amazon Nova Lite: The balanced workhorse for everyday professional tasks like drafting emails.

Amazon Nova Pro: A powerful multimodal model that handles complex instructions and data analysis with high accuracy.

Scout and Maverick

Meta’s latest open-weight models offer strong efficiency and logic.

Llama 4 Scout: A nimble model designed for speed and quick logical checks. It performs well above its weight class for its size.

Llama 4 Maverick: A highly capable and versatile model that excels at reasoning, following complex instructions, and general problem-solving.

Haiku, Sonnet, and Opus

The Claude family is known for nuanced understanding, safety, and sophisticated writing.

Claude 4.5 Haiku: A fast model suited to near-instant responses that still require strong reasoning.

Claude 4.6 Sonnet: Balances speed and high intelligence for a wide range of professional tasks.

Claude 4.6 Opus: The most powerful model in the lineup. Use it for complex tasks that require deep synthesis and creativity.

20B and 120B · Text only

OpenAI’s open-weight models provide strong text generation and reasoning capabilities. Both models work with text only.

OpenAI GPT-OSS 20B: A lean and efficient generalist for everyday drafting, summarizing, and quick Q&A.

OpenAI GPT-OSS 120B: Designed for multi-step analysis, complex instructions, and technical problem-solving.

Choose a starting tier

Compare the capabilities of each available model to find the best fit for your task.

Comparison of AI model tiers, available models, and recommended uses.
TierModelsBest For
Fast & Efficient
(Simple Tasks)
  • Amazon Nova Micro
  • Llama 4 Scout
Categorizing text, simple grammar checks, translating short phrases, or generating high-volume, low-complexity content.
The Workhorses
(Standard Tasks)
  • Amazon Nova Lite
  • Amazon Nova Pro
  • Claude 4.5 Haiku
  • OpenAI GPT-OSS 20B
Drafting reports, summarizing long meetings, creating meeting agendas, and general administrative assistance.
High Performance
(Advanced Tasks)
  • Claude 4.6 Sonnet
  • Llama 4 Maverick
  • OpenAI GPT-OSS 120B
  • Qwen3 32B
Complex coding, technical documentation, structured data extraction, and logical reasoning.
Frontier Intelligence
(Expert Tasks)
  • Claude 4.6 Opus
Strategic planning, nuanced creative writing, high-stakes synthesis, and deep analytical “thinking.”

Model Comparison Table

Compare the capabilities of each available model to find the best fit for your task.

Comparison of AI model capabilities and relative cost. One dollar sign means low relative cost, two dollar signs mean moderate relative cost, and three dollar signs mean high relative cost. A check mark means the capability is available, and a dash means it is not available.
ModelRelative CostLong ContextMultimodal (Images)Code GenerationFast ResponsesAdvanced Reasoning
Amazon Nova Micro
(Amazon Nova)
Low relative cost Not available Not available Not available Available Not available
Amazon Nova Lite
(Amazon Nova)
Low relative cost Not available Available Not available Available Not available
Amazon Nova Pro
(Amazon Nova)
Moderate relative cost Available Available Available Not available Available
Llama 4 Scout
(Meta)
Low relative cost Not available Not available Available Available Not available
Llama 4 Maverick
(Meta)
Low relative cost Available Not available Available Not available Available
Claude 4.5 Haiku
(Anthropic)
Moderate relative cost Not available Not available Available Available Available
Claude 4.6 Sonnet
(Anthropic)
High relative cost Available Available Available Not available Available
Claude 4.6 Opus
(Anthropic)
High relative cost Available Available Available Not available Available
Qwen3 32B
(Alibaba)
Low relative cost Available Not available Available Not available Available
OpenAI GPT-OSS 20B
(OpenAI, open-source)
Low relative cost Not available Not available Available Available Not available
OpenAI GPT-OSS 120B
(OpenAI, open-source)
Low relative cost Available Not available Available Not available Available

Which model should I use?

These common UCSB work and teaching scenarios show how task complexity translates into a model choice.
 

1

Drafting a Professional Email

Task

Draft a polite follow-up email to a colleague.

Recommended model

Amazon Nova Lite or Llama 4 Scout

Why

These models are quick and capable of handling standard professional tone and structure.
 

2

Summarizing a Research Paper

Task

Summarize a 20-page PDF into a three-paragraph summary and five bullet points.

Recommended model

Amazon Nova Pro or Claude 4.5 Haiku

Why

These models have larger context windows and stronger synthesis capabilities than Tier 1 models.
 

3

Writing a Python Script

Task

Write a Python script to merge data from multiple Excel files.

Recommended model

Claude 4.6 Sonnet or Llama 4 Maverick

Why

Coding requires high-level logical reasoning. Sonnet is a strong coding assistant, while Maverick provides robust open-weight logic.
 

4

Analyzing Survey Data

Task

Identify themes, sentiments, and actionable insights from hundreds of open-ended survey responses.

Recommended model

Claude 4.6 Opus

Why

This high-complexity task requires a nuanced understanding of human sentiment and its strategic implications.

5

Preparing Course Communications

Task

Turn a lecture outline into an announcement, three discussion prompts, and a short quiz recap.

Recommended model

OpenAI GPT-OSS 20B

Why

This is routine, high-frequency drafting. GPT-OSS 20B produces consistent instructional text with minimal overhead.
 

6

Building a Problem Set

Task

Create a 10-question problem set with increasing difficulty and a step-by-step worked answer key.

Recommended model

OpenAI GPT-OSS 120B

Why

Correct worked solutions require genuine multi-step reasoning. GPT-OSS 120B is designed for this kind of work.