Krishnapadala55/brahmastra-0.2

TEXT GENERATIONPricing:Input $2.72 / Output $4.8Concurrent Unit Cost:2Model Size:32.8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Apr 10, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Krishnapadala55/brahmastra-0.2 is a 32.8 billion parameter reasoning model based on DeepSeek-R1-Distill-Qwen-32B, specifically engineered for Dynamic Application Security Testing (DAST). This model excels at autonomously identifying web application vulnerabilities, generating targeted security payloads, analyzing HTTP responses, and producing structured security findings. It features a 6-phase training curriculum and explicit traces for step-by-step reasoning, making it highly effective for authorized security testing and penetration testing assistance.

Loading preview...

BRAHMASTRA 0.2: AI-Native DAST Security Scanner

BRAHMASTRA 0.2, developed by Krishnapadala, is a 32.8 billion parameter reasoning model built upon the DeepSeek-R1-Distill-Qwen-32B base. It is purpose-built for Dynamic Application Security Testing (DAST), designed to autonomously reason about web application vulnerabilities, generate security payloads, analyze HTTP responses, and identify authentication session drops.

Key Differentiators & Capabilities

This version represents a significant upgrade from its 7B predecessor, featuring:

  • Enhanced Reasoning: Explicit <think> traces enable detailed, step-by-step vulnerability analysis.
  • Expanded Training: A 6-phase training curriculum (p1a-p6) on diverse datasets like Fenrir v2.0, HackMentor, and ExploitDB, covering a wide array of attack types.
  • Comprehensive Vulnerability Coverage: Trained on 28 "Astra modules" covering critical, high, and medium severity vulnerabilities, including SQL/NoSQL Injection, SSTI+RCE, Insecure Deserialization, SSRF, XSS, IDOR, JWT/OAuth bypasses, Path Traversal, GraphQL attacks, WAF detection & bypass, and business logic flaws.
  • Optimized Deployment: While the base model requires a 48 GB GPU, quantized versions (Q4_K_M) can fit on GPUs with ~20 GB VRAM, making it accessible for various setups.

Intended Use Cases

BRAHMASTRA 0.2 is designed for authorized security testing only and is ideal for:

  • DAST scanner backends for payload generation and response analysis.
  • Penetration testing assistants for red teams.
  • Bug bounty triage and reproduction scripting.
  • Security training platforms and CTF solvers.
  • Vulnerability research and exploit chain modeling.

Limitations

Users should be aware of potential reasoning verbosity due to <think> blocks, possible hallucination on obscure CVEs (post Q1 2026), and the need to verify blind-injection payloads due to potential false positives from WAF blocks. The training corpus is primarily English, which may affect non-English targets.

Finetunes
2 models