AI Tools Reviewed

Free Models for AI Development

‍

Meta (Llama)

• Speed: Average
• Accuracy: Average but dependent on your training
• File Search/Upload?: Yes (if built)
• Premium Level: No
• Security: Dependent on your setup

‍

Meta's Llama is arguably the best free model available for local development, with several cloud-hosted options. While the model is fast, it lacks accuracy compared to OpenAI and requires significant training to improve performance. Fortunately, the development on top of Llama is becoming easier with a plethora of tools and guides being released. If you’re considering a custom solution that doesn’t rely on sending information to a cloud-based model like OpenAI or Anthropic, Llama is likely your best bet.

‍

“Freemium” Models

‍

ChatGPT by OpenAI

• Speed: Fast
• Accuracy: Below average to average
• File Search/Upload?: Yes
• Premium Level: Yes
• Security: SOC2

‍

ChatGPT is the most pervasive and well-known model, but it falls short for business use cases due to its lack of accuracy. Large blocks of text often require additional editing, coding tasks can be repetitive and error-prone, and the file search uses a middle-out algorithm that can skip important pieces. OpenAI’s “one model fits all” approach has led to a decline in quality and accuracy since the launch of GPT-4.

‍

Despite its powerful features like indexing and vector-stores off the shelf, the lack of accuracy precludes its use in business applications at scale. The model limits and context windows also tend to shift without notice, frustrating developers. While enterprise plans and paid resources might offer more consistent and higher-quality performance, they come at a custom cost. OpenAI's strength lies in its fast iteration and frequent launches of new models, but this speed has come at the expense of quality.

‍

Anthropic

• Speed: Fast
• Accuracy: Above average (especially for coding tasks)
• File Search/Upload?: Yes
• Premium Level: Yes
• Security: SOC2

‍

Anthropic’s Claude Sonnet, Haiku, and Opus models are direct competitors to OpenAI, offering a variety of UI and API-based usage scenarios for businesses. The models tend to be more accurate, but they come with a lower context window and limitations on daily usage. These limitations can be frustrating for businesses requiring high throughput for document or larger text/image processing. However, once you get comfortable with their API, it is very consistent in its responses.

These freemium models excel at basic tasks like blog/marketing generation, simple document reviews, and math-based tasks but often struggle with more complex or varied tasks.

‍

Google (Gemini)

• Speed: Fastest
• Accuracy: Above average (especially for text-based tasks)
• File Search/Upload?: Yes
• Premium Level: Yes
• Security: SOC

‍

Google Gemini, with its Flash and Pro models, offers the largest context windows available (over 2.8M tokens on Flash) and blazing fast response times. Well supported and documented by Google, their multi-model techniques have significantly improved. We have been impressed enough with Google to integrate some of their modeling into our internal processes. The speed and accuracy are highly usable at scale, making it our recommended model for small to medium-sized businesses looking to leverage AI.

Use the API versions of these models, as they will require some development experience.

‍

‍

A Quick Note on Enterprise Models

Enterprise models from companies like Salesforce, IBM, and others are available at a higher cost and are highly trained on your datasets. These are often significant in both costs and setup but can be better suited for organizations with very defined processes or those already within a relevant ecosystem.

‍