Deep Cogito Unveils Hybrid AI Models with Advanced Reasoning Capabilities
Deep Cogito, a new AI innovator, has launched a series of openly accessible AI models that seamlessly switch between reasoning and non-reasoning modes.
Reasoning models, such as OpenAI’s o1, excel in fields like mathematics and physics by methodically verifying solutions step by step. However, this precision demands greater computational resources and time. To address this, companies like Anthropic are developing hybrid architectures that blend reasoning with standard components, enabling quick responses to simple queries and deeper analysis for complex ones.
Deep Cogito’s models, named Cogito 1, are all hybrid designs. The company claims they surpass top open models of similar size, including those from Meta and DeepSeek.
“Each model can respond instantly or reflect before answering, mimicking reasoning models,” Deep Cogito stated in a blog post. “They were crafted by a compact team in just 75 days.”
Cogito 1 models range from 3 billion to 70 billion parameters, with plans for models up to 671 billion parameters in the near future. More parameters typically enhance a model’s problem-solving prowess.
Cogito 1 builds on Meta’s Llama and Alibaba’s Qwen models, with Deep Cogito applying innovative training techniques to improve performance and enable switchable reasoning.
Internal benchmarks show the largest model, Cogito 70B, with reasoning enabled, outperforms DeepSeek’s R1 on select math and language tasks. Without reasoning, it also surpasses Meta’s Llama 4 Scout on LiveBench, a broad AI evaluation.
All Cogito 1 models are available for download or accessible via APIs on Fireworks AI and Together AI cloud platforms.

Cogito 1’s performance compared to other popular openly available AI modelsImage Credits:Deep Cogito “We’re still early in our scaling journey, using only a fraction of the compute typically allocated for large language model training,” Deep Cogito noted in its blog. “We’re exploring new post-training methods for self-improvement.”
Founded in June 2024 in San Francisco, per California filings, Deep Cogito was established by Drishan Arora, a former Google senior software engineer, and Dhruv Malhotra, previously a product manager at Google’s DeepMind, where he focused on generative search.
Backed by South Park Commons, according to PitchBook, Deep Cogito aims to pioneer “general superintelligence”—AI that outperforms most humans and unlocks unprecedented capabilities.
Related article
Stripe to Acquire OpenRouter as AI Model Routing Gains Momentum
Stripe has agreed to acquire OpenRouter, an AI model-routing platform that gives developers access to hundreds of models through a single interface. The deal adds model selection and routing to Stripe’s existing work around AI usage and token-based b
Former OpenAI Researcher Barrett Zoph Returns to Google to Support Gemini Development
Barret Zoph, who previously abruptly left and joined OpenAI due to a disagreement with Thinking Machines Lab CEO Miriam Mularati, has once again changed employers. Alphabet's Google announced that Zoph will return as Vice President of Research, bring
How Recursive Intelligence Raised $335M at $4B Valuation in 4 Months
The co-founders of startup Ricursive Intelligence seemed destined to be co-founders.Anna Goldie, CEO, and Azalia Mirhoseini, CTO, are so well-known in the AI community that they were among those AI engineers who “got those weird emails from Zuckerber
Related Special Topic Recommendations
Comments (0)
0/500
Deep Cogito, a new AI innovator, has launched a series of openly accessible AI models that seamlessly switch between reasoning and non-reasoning modes.
Reasoning models, such as OpenAI’s o1, excel in fields like mathematics and physics by methodically verifying solutions step by step. However, this precision demands greater computational resources and time. To address this, companies like Anthropic are developing hybrid architectures that blend reasoning with standard components, enabling quick responses to simple queries and deeper analysis for complex ones.
Deep Cogito’s models, named Cogito 1, are all hybrid designs. The company claims they surpass top open models of similar size, including those from Meta and DeepSeek.
“Each model can respond instantly or reflect before answering, mimicking reasoning models,” Deep Cogito stated in a blog post. “They were crafted by a compact team in just 75 days.”
Cogito 1 models range from 3 billion to 70 billion parameters, with plans for models up to 671 billion parameters in the near future. More parameters typically enhance a model’s problem-solving prowess.
Cogito 1 builds on Meta’s Llama and Alibaba’s Qwen models, with Deep Cogito applying innovative training techniques to improve performance and enable switchable reasoning.
Internal benchmarks show the largest model, Cogito 70B, with reasoning enabled, outperforms DeepSeek’s R1 on select math and language tasks. Without reasoning, it also surpasses Meta’s Llama 4 Scout on LiveBench, a broad AI evaluation.
All Cogito 1 models are available for download or accessible via APIs on Fireworks AI and Together AI cloud platforms.

“We’re still early in our scaling journey, using only a fraction of the compute typically allocated for large language model training,” Deep Cogito noted in its blog. “We’re exploring new post-training methods for self-improvement.”
Founded in June 2024 in San Francisco, per California filings, Deep Cogito was established by Drishan Arora, a former Google senior software engineer, and Dhruv Malhotra, previously a product manager at Google’s DeepMind, where he focused on generative search.
Backed by South Park Commons, according to PitchBook, Deep Cogito aims to pioneer “general superintelligence”—AI that outperforms most humans and unlocks unprecedented capabilities.
Stripe to Acquire OpenRouter as AI Model Routing Gains Momentum
Stripe has agreed to acquire OpenRouter, an AI model-routing platform that gives developers access to hundreds of models through a single interface. The deal adds model selection and routing to Stripe’s existing work around AI usage and token-based b
Former OpenAI Researcher Barrett Zoph Returns to Google to Support Gemini Development
Barret Zoph, who previously abruptly left and joined OpenAI due to a disagreement with Thinking Machines Lab CEO Miriam Mularati, has once again changed employers. Alphabet's Google announced that Zoph will return as Vice President of Research, bring
How Recursive Intelligence Raised $335M at $4B Valuation in 4 Months
The co-founders of startup Ricursive Intelligence seemed destined to be co-founders.Anna Goldie, CEO, and Azalia Mirhoseini, CTO, are so well-known in the AI community that they were among those AI engineers who “got those weird emails from Zuckerber





Home






