Model comparison
train_llama8b_grpo_single_dartmath_fromsft vs o3
Capability radar, value bars, and side-by-side posture.
train_llama8b_grpo_single_dartmath_fromsft
Not scored
value score
o3
93
value score
train_llama8b_grpo_single_dartmath_fromsft ctx
—
o3 ctx
200K
Capability radar
Value · context · multimodal · openness · speed posture
Value head-to-head
- o393
Attribute tape
| Field | train_llama8b_grpo_single_dartmath_fromsft | o3 |
|---|---|---|
| Developer | Unknown | OpenAI |
| Context | — | 200K |
| Modalities | Text, Code | |
| Openness | — | Proprietary Model |
| Speed | — | — |
| Price | — | $10/1M in · $40/1M out |
| Value score | — | 93 |
| Summary | A community fine-tuned Llama 8B model checkpoint trained with GRPO reinforcement-learning optimization on top of an SFT base, targeting math and reasoning tasks via a dartmath-style dataset. Released as an individual experiment on Hugging Face by user YukinoshitaYukino. | OpenAI reasoning model focused on hard problems. |