Cogito v2 Preview Llama 405B
Release in the Cogito family · version cogito-v2-preview-llama-405B
Maintained by Deep Cogito14
A 405B-parameter dense hybrid reasoning model that Deep Cogito post-trained from Meta's Llama 3.1 405B, released in July 2025 as one of four Cogito v2 Preview models. It can answer directly or reason first, supports a 128k-token context, and was trained in over 30 languages using Deep Cogito's Iterated Distillation and Amplification approach.145
- Model hub: Model card (Hugging Face) (external site: huggingface.co)
- Release notes: Cogito v2 Preview announcement (external site: deepcogito.com)
- Repository: Cogito v2 and v2.1 repository (external site: github.com)
Availability and license
Overall availability
Weights download from Hugging Face without an access gate; use is subject to the Llama 3.1 Community License. Deep Cogito's repository also lists hosted access through OpenRouter and Together AI.125
Availability is separate from permission: read the license before using or redistributing.
The model card's license field is "llama3.1" and its license section says the repository and weights are under the Llama 3.1 Community License Agreement. That agreement requires including it and displaying "Built with Llama" when redistributing, beginning the names of distributed derived models with "Llama", and following Meta's acceptable use policy; it requires licensees whose products had more than 700 million monthly active users in the month before the Llama 3.1 release to request a license from Meta, and ends for anyone who sues Meta or any other entity alleging that the Llama materials or outputs infringe their IP. California law governs.16
Model-disclosure tier
The model parameters for this release can be downloaded by the public. License terms may still restrict use, redistribution, or commercial use.
The weights are under a license that is not on the rubric's OSI-approved list. Read its terms before use.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| WeightsCan the general public download the model parameters for this release? | Public | BF16 weights in an ungated Hugging Face repository.12 |
| Inference codeIs code for running the model published? | Public | The model card gives Transformers examples for standard and reasoning modes and for tool calling; the repository's config declares the LlamaForCausalLM architecture.13 |
| Training codeIs the code used to train the model published? | Unknown | No training code was found in the model card, announcement, or GitHub repository. |
| Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access. | Unknown | The model card and announcement do not describe the post-training data; Meta's pretraining data for the base model is not public. |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Partial | The announcement describes Iterated Distillation and Amplification (distilling the results of inference-time reasoning back into the model's parameters) at a conceptual level without configurations or data details.41 |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Partial | The model card shows a benchmark chart and refers to the announcement for more results; evaluation code is not linked.14 |
What it is useful for
The model card describes it as optimized for coding, STEM, instruction following, and general helpfulness, with multilingual, coding, and tool-calling abilities.1
Run and use notes
- The model card shows Transformers usage in bfloat16 and explains that reasoning is enabled by passing enable_thinking=True to the chat template or by a specific system prompt with the response prefilled with a think tag.1
Organization context
Provenance and derivatives
Post-trained by Deep Cogito from Meta's Llama 3.1 405B base model. Deep Cogito's contribution is the post-training; the base model was pretrained by Meta.15
- Derived from: Llama 3.1 405B (meta-llama/Llama-3.1-405B) — Meta base model; see the Llama family record.
Other releases in the Cogito family
- Cogito v2.1 671BModel-disclosure tier (USASI rubric v0.1): Open-weight
U.S. eligibility
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.