Cogito

Cogito is an innovative set of language models created by Deep Cogito, a startup in San Francisco. The company was started by former DeepMind and Google researchers. Their goal is to change AI with advanced thinking abilities.

Benefits

Cogito models have several key benefits. They can switch between standard and thinking modes, balancing speed and efficiency. This makes them very versatile and powerful. The main 70B model does better than other top models in areas like math reasoning and language understanding. Cogito''s special training method, called Iterated Distillation and Amplification (IDA), lets the models improve through self improvement cycles. This makes them more scalable and time efficient. The 70B model, for example, was made in just 75 days.

Use Cases

Cogito models are great for many tasks, like coding, function calling, and agent tasks. They can work in both standard and thinking modes, making them useful for many applications. They are not made for very long thinking tasks. The models support a 128,000 token context window and strong multilingual abilities across 30 languages. They also have built in tool calling features, letting applications get data, run calculations, or send emails with one command.

Pricing

The models are available for download or use via APIs on platforms like Huggingface, Ollama, Fireworks AI, and Together AI. The open license lets companies add the models to paid products for up to 700 million users each month without extra fees.

Vibes

Early testers liked Cogito''s advanced features and abilities. The models'' large context window, strong multilingual support, and built in tool calling got good feedback.

Additional Information

Deep Cogito plans to release even larger Mixture of Experts (MoE) models with sizes reaching 109B, 400B, and 671B. These will be released under an open license, matching the company''s mission to reach general superintelligence through advanced reasoning and iterative self improvement.