Making Sense of OpenAI’s Models

April 22, 2025 

A few quick notes from the intersection of AI, marketing, branding, and small business.

New thinking, this time with images. OpenAI released a number of new models last week. Many commentators have been effusive about the new capabilities, including what OpenAI calls “thinking with images.” Here’s an article from The New York Times and another from MacRumors.

However, confusion reigns. I don’t know about you, but I’ve had a hard time keeping track of all the new offerings from OpenAI, and what each offering is best for. Clearly OpenAI needs to hire a naming expert to clear things up (ahem, TalktoZilly). So in the spirit of understanding, here are the models currently available from OpenAI (as of April 22, 2025), including use cases, strengths, and limitations. Costs and who has access to what are at the bottom of this post.

GPT-4o

• Description: OpenAI’s flagship model, natively multimodal (text, vision, audio), faster and more cost-effective than GPT-4, and now the default for most ChatGPT users14.

• Use Cases:

  • Real-time multimodal interactions (analyzing images, audio, and text)

  • Advanced customer support with voice and vision

  • Creative collaboration (storytelling, brainstorming, multimedia content)

  • Coding (frontend/backend, debugging, code review)

  • STEM problem-solving and academic research

  • Knowledge discovery using up-to-date web search145

• Strengths: Fast, creative, understands implied intent, excels at instruction following, supports all tools (web search, data analysis, file uploads, image generation).

• Limitations: Slightly different “personality” and output style compared to GPT-4; some users find it more verbose3.

GPT-4o Mini

• Description: Lightweight, fastest, and most cost-effective version of GPT-4o; available to all users and used automatically when GPT-4o rate limits are reached4.

• Use Cases:

  • Everyday, low-stakes tasks (summarizing, quick research, basic Q&A)

  • High-volume backend processing (sentiment analysis, topic modeling, data sorting)

  • Content generation for blogs, outlines, and drafts

  • Summarizing long transcriptions or social media analysis at scale34

• Strengths: Extremely fast and cheap, ideal for automation and background tasks.

• Limitations: Does not support advanced tools (data analysis, file uploads), weaker at vision tasks, lower reasoning depth than full GPT-4o.

o-Series (o3, o4-mini, etc.)

• Description: Latest experimental models focused on advanced reasoning and agentic task execution; can autonomously use all ChatGPT tools and combine them to solve complex, multi-step problems1.

  • Use Cases:

  • Multi-faceted problem-solving (combining search, coding, data analysis, and image generation)

  • Research requiring synthesis from multiple sources

  • Tasks needing autonomous execution (e.g., planning, workflow automation)

  • Advanced academic and enterprise applications1

• Strengths: State-of-the-art reasoning, can independently decide which tools to use, handles complex queries with multiple steps.

• Limitations: Newest models—may have evolving capabilities and availability.

Anyone can use the newest and most powerful o3 model, but there are some limitations and costs depending on how you want to access it:

ChatGPT Plus, Pro, and Team subscribers: As of April 16, 2025, o3 is available to all paying ChatGPT Plus, Pro, and Team users within the ChatGPT interface. Pro subscribers have almost unlimited access, while Plus and Team users also get priority access to o3 and o4-mini. These models are replacing the older o1 and o3-mini options in ChatGPT.

ChatGPT Free users: Free-tier users do not have direct access to o3. However, they can try out o4-mini, a related but less powerful model, through the 'Think' option in ChatGPT.

API access for developers: o3 is available via the OpenAI API at a rate of $10 per million input tokens and $40 per million output tokens, which is higher than previous models due to its advanced capabilities. This makes it accessible to anyone willing to pay for API usage, but the cost may be significant for large-scale or high-compute tasks.

Third-party platforms: Some platforms, like Writingmate, offer access to o-series models (including o3) as part of their subscription services, which can be a more cost-effective option for some users.