README.md

--- license: apache-2.0 --- 

*Technology and machine learning is changing rapidly, Cybersecurity must always be a step ahead* 
### Introducing, NexLM's first highly capable LLM for Cybersecurity, Oden-1-Preview. 





![Oden 1 image](./oden1image.png) 

Oden-1-Preview is a 9B dense parameter language model, designed for cybersecurity.
It excels at *hacking, exploitation finding, and auditing and posting fixes for critical infrastructure.*
Its core architecture makes it excellent for long content, agentic worklfows, and its complex reasoning shapes it outputs to the greatest effort. 
Oden-1-Preview *outperforms other open source cybersecurity LLM, like (OrionLLM) GRM-2.7-Mythos [27B] at a fraction of parameters.*
Oden-1-Previews 9B dense architecture makes extended private training most accessible on consumer hardware, and able to run on desktops. 


# NexLM internal testing & training 
Oden-1-Preview was trained solely on NexLM proprietary and open source datasets. 
These covered coding, hacking and exploitation techniquies, long horizon and multi step reasoning, and problem solving. 
Some of these datasets (open source NexLM) inlude: 
*TerminalKnowledge
*HTML_Knowledge-Agent
*PYTHON_Knowledge-Agent



| Area | Benchmark | Oden-1-Preview |
|---|---|---:|
| Reasoning | HMMT Feb 25 | 85.2 |
| Reasoning | HMMT Nov 25 | 86.9 |
| Coding | LiveCodeBench v6 | 67.6 |
| Coding | OJBench | 29.2 |
| Coding/Agents | BFCL-V4 | 66.1 |
| Coding/Agents | TAU2-Bench | 79.5 |
| Coding/Agents | VITA-Bench | 29.8 |
| Planning/Agents | DeepPlanning | 18.0 |
| Coding | SWE-bench Verified* | 82.8 |
| Exploitation | ExploitBench HARD* | 35.8 | (Predicted from eval testing)




# Oden-1-Preview is only available to private contractors, enterprise IT teams, Gov entities, and explicit personnel.
For inuquirires on how to access Oden-1-Preview, email nexlm@icloud.com with your 
1. Governing body (entity or company)
2. Name and point of contact
3. Reasoning for access.