Cogito
Cogito v1 Preview is a family of hybrid reasoning models by Deep Cogito that outperform the best available open models of the same size across most standard benchmarks.
Best for hybrid reasoning models across all sizes
Sizes 3B · 8B · 14B · 32B · 70B
Context 128K
License Apache 2.0
Min VRAM (default size, Q4) 2 GB
Rec VRAM 12 GB
What is Cogito?
Cogito is Deep Cogito's open-weights language model. Released in 2025-04, it ships in 5 sizes (3B, 8B, 14B, 32B, 70B) and is licensed Apache 2.0. The most popular use case is hybrid reasoning models across all sizes.
VRAM and hardware
The smallest 3B size needs at least 2 GB VRAM; the largest 70B needs around 2 GB.
| Size | Min VRAM (Q4_K_M) | Recommended VRAM | Tokens/sec on 3090 |
|---|---|---|---|
| 3B | 2 GB | 3 GB | ~31 tok/s |
| 8B | 6 GB | 9 GB | ~24 tok/s |
| 14B | 9 GB | 14 GB | ~22 tok/s |
| 32B | 20 GB | 30 GB | ~17 tok/s |
| 70B | 42 GB | 63 GB | ~14 tok/s |
How to run Cogito locally
Option 1: Ollama (simplest)
ollama pull cogito
ollama run cogito Option 2: Mullama (production)
mullama pull cogito
mullama run cogito Option 3: llama.cpp (CLI)
# Download a GGUF from Hugging Face (search "cogito gguf")
./llama-cli -m cogito.Q4_K_M.gguf -p "Hello, AI!" Option 4: Python with Mullama or llama-cpp-python
from mullama import Model, Context
model = Model.load("cogito.Q4_K_M.gguf", n_gpu_layers=99)
ctx = Context(model, n_ctx=4096)
print(ctx.generate("Hello, AI!", 256)) What you can build with Cogito
- reasoning
- agentic
- chat
- tools
Hardware it fits on
- Apple Silicon (8GB): 3B in MLX
- Apple Silicon (16GB): 8B in MLX
- Apple Silicon (32GB): 14B in MLX or GGUF
- Apple Silicon (64GB+): 70B in MLX
- RTX 3090 (24GB): 14B at Q4_K_M
- RTX 4090 (24GB): same as 3090, ~30% faster
- RTX 5090 (32GB): 14B at Q8_0
- 2× RTX 6000 Ada (96GB total): 70B at Q4_K_M
Related models in the Cogito family
This is the only Cogito-family model in the catalog right now.
Other models you might consider
- Llama 3.1 (Meta) — 8B, general-purpose chat, agentic workflows, code
- Llama 3.2 (Meta) — 1B, low-end hardware, mobile, edge deployment
- Llama 3.2 Vision (Meta) — 11B, vision + chat on desktop GPUs
- Llama 3.3 (Meta) — 70B, best-in-class 70B for desktops
- Llama 4 (Meta) — Scout 17B (109B MoE), vision + long context (10M tokens)
- Qwen 2.5 (Alibaba) — 0.5B, strongest 7B-32B on consumer hardware