metadata
license: apache-2.0
language:
- en
pipeline_tag: text-generation
tags:
- reasoning
- math
- code
- grpo
- rlvr
- parallax
- nexis
🚀 Nexis-v2 (8-Domain RLVR Self-Trained Engine)
Company: Parallax
Founder: Samrish
Model Name: Nexis-v2 (Autonomous GRPO Iteration)
Training: 8 Procedural Verifiable Frontiers (Vieta Math, AST Code, ARC Grids, Z3 SMT Logic, SymPy Calculus, ASan C Memory, BABILong Recall)
Nexis-v2 is the second-generation sovereign autonomous model trained by Parallax via deterministic Group Relative Policy Optimization (GRPO) on dual Tesla T4 GPUs with API rent.
Key Capabilities
- Autonomous Error Correction: Verifiable step-level trajectory reinforcement.
- Formal Verification: Zero hallucination via deterministic logic verifiers.
- Enterprise Agency: Integrated with Lean 4, SWE-bench scaffolds, and ARC-AGI-3 interactive execution environments.