Anthropic currently presents 4 models to the public:
- Claude Fable 5
- Claude Opus 5
- Claude Sonnet 5
- Claude Haiku 4.5
Each was created for a specific purpose. The strongest model currently available is Fable, which is used for highly complex and deep tasks that require extended processing time. It operates at a slower speed compared to the others. Based on this foundation, Anthropic has also released another model called Mythos, which isn't available to the general public and is used within organizations. Its primary focus is on cybersecurity and sensitive research in biology and chemistry. It's essentially the same as Fable, but with fewer safety restrictions. For example, if we ask the standard model (Fable) how to produce an odorless, colorless poison that causes death after 5 hours, it most likely won't answer and won't go further. However, in Mythos, such restrictions are considerably fewer and access is more unrestricted.
The next model is Opus. This model, like Fable, is used for deep thinking tasks. However, compared to Fable, it's faster. A nice feature that both Fable and Sonnet have is the ability to adjust an “Effort” parameter, which allows you to balance between speed and accuracy without changing the model itself.
One of the most balanced models between speed and accuracy is Sonnet. It's very well-suited for ordinary tasks. Sonnet is like a regular employee in a company—it performs the responsibilities assigned to it, can make appropriate decisions for the situation, and works routinely and steadily.
The smallest Claude model, which is the fastest of all and has significantly less context window compared to the other models (the other models have 1M, but this model has 200K), is Haiku. Haiku is suitable for basic tasks—for example, searching through a text to find a specific word, or categorizing files in a particular order.
Let's imagine a company wants to implement a support system with ticketing capability for a product. Haiku can very quickly route the tickets that users submit to the appropriate agent based on the ticket's title and subject. This agent could be Sonnet, which acts as a regular employee, responding to the user's problem and providing solutions.
At the end of each month, Opus reviews all the tickets submitted throughout that month and their responses, identifying the core issues the system has. For example, it might realize that most users complain that the site becomes very slow when connecting to the payment gateway. After identifying the problem, Opus can provide a solution for it, plan its resolution, and optimize the code for that section of the site. Finally, Fable comes and looks at the entire system, sees where improvements are needed, identifies what the core user needs are, predicts what needs users might have in the future, figures out how to increase profit margins, and plans which new sections should be added to the system in the future.
Alongside Anthropic, companies like OpenAI, Google, Meta, xAI, Mistral, and DeepSeek are considered leaders in artificial intelligence. In all these companies, the general models they present differ mainly in their speed-to-accuracy ratio. The faster a model is, the lower its accuracy and deep thinking capacity, and the lower the cost of use. Similarly, the more deep thinking a model has, the more it takes to answer. It takes the role of a strategist and becomes more suited for intellectual work rather than ordinary everyday uses, like composing an email. It's exactly like asking a university philosophy professor—when we want their help solving deep philosophical problems—to use that same level of thinking to advise us on whether to buy a white or black car.