A new company, Deep cogitoIt has emerged from Stealth with a family of openly available AI models that can be converted between ways of “reasoning” and non -disappearance.
Models of reasoning such as O1 of Openai have shown great promise in areas such as mathematics and physics, thanks to their ability to effectively control themselves, worked through complex step -by -step problems. This reasoning comes at a cost, however: a higher computational and latent situation. That is why workshops such as Anthropic are followed by “hybrid” models that combine logic ingredients with standard, non-eventing elements. Hybrid models can quickly answer simple questions while spending extra time taking into account more provocative questions.
All Deep Cogito models, called cogito 1, are hybrid models. Cogito claims to exceed the best open models of the same size, including Meta and Chinese AI Startup Deepseek models.
“Every model can answer directly […] or self-reflect before answering (such as models of reasoning), “the company Explained to a blog post. “[All] were developed by a small group in about 75 days. ”
Cogito 1 models range from 3 billion parameters to 70 billion parameters and Cogito reports that models ranging from up to 671 billion parameters will participate in the coming weeks and months. The parameters correspond approximately the problem solving skills of a model, with most parameters generally better.
Cogito 1 was not developed from scratch to be clear. Deep Cogito was built over the models of Meta Open Llama and Alibaba’s Qwen to create its own. The company says it has applied new training approaches to enhance the performance of the base models and to activate reasoning.
According to the results of Cogito’s internal comparative assessment, the largest Cogito 1 model, Cogito 70B, with logic surpassed Deepseek R1 Reasoning Model in some mathematical and linguistic evaluations. The Cogito 70B with the disabled Eclipses Eclipses Meta recently released the Llama 4 Scout model in Livebench, a general AI test.
Each Cogito 1 model is available for download or use via API in cloud fireworks AI and AI.
‘At present, we are still in the early stages of [our] The escalation curve, having only used one fraction of calculation that is usually intended for traditional linguistic publication/continuing training, “Cogito wrote in place on her blog.
According to archiving with the state of CaliforniaDeep Cogito based in San Francisco was founded in June 2024. The company LinkedIn Page It quotes two co -founders, Drissan Arora and Dhruv Malhotra. Malhotra was previously a product manager at Google Ai Lab Deepmind, where she worked in genetic search technology. Arora was a superior software engineer on Google.
Deep Cogito, whose supporters include South Park Commons, According to PitchbookIt is ambitiously aimed at building the “general hyperthy”. The founders of the company understand the phrase which means AI that can perform better than most people and “reveal completely new possibilities we have not yet imagined”.
