Model comparison
pick_cube_lora_1.8w vs GPT-5
Capability radar, value bars, and side-by-side posture.
pick_cube_lora_1.8w
Not scored
value score
GPT-5
94
value score
pick_cube_lora_1.8w ctx
—
GPT-5 ctx
400K
Capability radar
Value · context · multimodal · openness · speed posture
Value head-to-head
- GPT-594
Attribute tape
| Field | pick_cube_lora_1.8w | GPT-5 |
|---|---|---|
| Developer | KaiyueChen | OpenAI |
| Context | — | 400K |
| Modalities | Text, Code, Image, Audio | |
| Openness | — | Proprietary Model |
| Speed | — | Fast |
| Price | — | Mid-High |
| Value score | — | 94 |
| Summary | A LoRA (Low-Rank Adaptation) fine-tuning adapter checkpoint published by KaiyueChen on Hugging Face, targeting a 'pick cube' robotic manipulation task. The adapter adds task-specific weights on top of a frozen base model using parameter-efficient fine-tuning, with approximately 1.8 thousand training steps indicated by the suffix. | OpenAI frontier multimodal model. |