OpenAI has introduced the o1-preview series of reasoning models designed to tackle complex problems in science, coding, and math. These models emphasize deeper reasoning before responding, achieving significant performance improvements over previous models. The o1 series includes enhanced safety measures and is particularly useful for complex tasks. Additionally, OpenAI is releasing o1-mini, a cost-effective model optimized for coding. Access to these models is available to ChatGPT Plus and Team users, with plans for broader access and feature enhancements in the future.
TechCrunch noted that while they excel at complex problem-solving, they are significantly more expensive and less effective for simpler tasks compared to GPT-4o. The model’s multi-step reasoning is promising for tackling big questions but often overcomplicates simpler inquiries.
OneUsefulThing noted that while it significantly improves reasoning, it still has limitations, including errors and hallucinations, and relies on the underlying GPT-4o model.
Big Technology did a deep dive – I considered including it in Big Ideas –so let me quote. “People who use AI for writing, editing, and marketing tasks will likely be disappointed. But people that use it for coding, math, and science research will be thrilled. In OpenAI’s testing, people who used o1 for writing actually preferred it less than GPT-4o. But those who used it for mathematical calculation, data analysis, and computer programming preferred it by a wide margin.”
OpenAI’s ChatGPT has surpassed 11 million paying subscribers, generating over $225 million in monthly revenue, according to COO Brad Lightcap. This includes 10 million standard subscribers and 1 million on higher-priced business plans, with users employing the service for tasks like coding, writing, and research.
There’s a potential bifrication of AI models here within a single product line that is already notable, and might even result in a split of MSP versus their customer. Customers may want the primary model, and technical use cases for managed IT will use the reasoning model. Until they merge at some point in the future, or the product guides to the correct model.
Instead, I think the key insight is that my sommelier analogy remains strong. Like an expert recommending a wine with dinner, a solution provider matches product (and embedded model) to a customer use case. Example. The fact that o1-preview is less favored for tasks like writing and marketing but excels at coding and math aligns with the core services MSPs typically offer. MSPs should assess when it’s worth deploying these models for costlier, complex projects, versus sticking with more general models for everyday operations.