Terminus-v2-1.2b-0356

1. Introduction

We are introducing the Terminus-v2 family, a series of reasoning models designed specifically for agentic coding within the Terminus-2 harness. These models are optimized for structured reasoning, task planning, logical decision-making, and organized tool use.

2. About this Checkpoint

Due to internal development constraints, this checkpoint was trained to only 4.45% of the planned training schedule.

Progress: [█░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░] 4.45%

Training will continue in future releases as we work toward completing the full model.

3. Benchmarks

Despite its early training stage, this checkpoint already demonstrates competitive performance on Terminal Bench 2.0.

Benchmark Terminus-v2-1.2b-0355 LFM2.5-1.2b-Thinking tmax-2b
Terminal Bench 2.0 (%) 3.2 0.0 2.9

Orion Research • 2026

Downloads last month
323
Safetensors
Model size
1B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for OrionLLM/Terminus-v2-1.2b-0356

Finetuned
(40)
this model
Quantizations
1 model