OrionLLM/Terminus-v2-1.2b-0356
OrionLLM/Terminus-v2-1.2b-0356 is a 1.2 billion parameter reasoning model from the Terminus-v2 family, developed by Orion Research. Designed for agentic coding within the Terminus-2 harness, it is optimized for structured reasoning, task planning, logical decision-making, and organized tool use. Despite being an early checkpoint trained to only 4.45% of its planned schedule, it shows competitive performance on Terminal Bench 2.0.
Loading preview...
Terminus-v2-1.2b-0356: An Agentic Reasoning Model
Orion Research introduces the Terminus-v2 family, a series of 1.2 billion parameter models specifically engineered for agentic coding. This particular checkpoint, Terminus-v2-1.2b-0356, is designed to excel within the Terminus-2 harness by focusing on structured reasoning and organized tool use.
Key Capabilities
- Agentic Coding: Optimized for tasks requiring autonomous decision-making and code generation.
- Structured Reasoning: Built for logical problem-solving and complex task planning.
- Tool Use: Designed to integrate and utilize external tools effectively.
Development Status and Performance
This model is an early development checkpoint, having completed only 4.45% of its intended training schedule. Despite its nascent stage, it demonstrates promising capabilities:
- Competitive Benchmarking: Achieves 3.2% on Terminal Bench 2.0, outperforming LFM2.5-1.2b-Thinking (0.0%) and tmax-2b (2.9%) at a similar parameter count.
Future releases are expected to continue training to unlock the model's full potential. This checkpoint offers an early look at a model focused on advanced reasoning for coding agents.