Models · Compare
Phi-3 Mini 4K Instruct vs GPT-4 Turbo
Rows highlighted in warm gray are where the models differ. Numbers carry their as-of date and primary source.
Specs
| Field | A: Phi-3 Mini 4K Instruct | B: GPT-4 Turbo |
|---|---|---|
| Released | 2024-04-23 | 2024-04-09 |
| Developer | Microsoft | OpenAI |
| Openness | Open | Proprietary |
| License | MIT | Proprietary |
| OSI-approved | yes | no |
| Data released | no | no |
| Training code | no | no |
| Architecture | dense | unknown |
| Total params | 3.8B | — |
| Active params | — | — |
| Experts | — | — |
| Context window | 4K | 128K |
| Attention | mha | unknown |
| Position enc. | rope | unknown |
| Pretraining tokens | 3.3T | — |
| Post-training | sft, dpo | rlhf |
| Training hardware | H100 | — |
| $/M input | — | $10.00 |
| $/M output | — | $30.00 |
| Output tok/sec | — | 27.8 |
Benchmarks
Missing scores render as not reported; never inferred. Bold highlights the leader per benchmark.
General reasoning
| MMLU | 70.9 2024-04-23 | — |
| MMLU-Pro | — | 69.4 2026-05-21 |
| GPQA-Diamond | 30.6 2024-04-23 | — |
Code
| HumanEval | 57.3 2024-04-23 | — |
| LiveCodeBench | — | 29.1 2026-05-21 |
Math
| MATH | — | 73.7 2026-05-21 |
| AIME 2024 | — | 15.0 2026-05-21 |
Context · A
Microsoft's first small-model release demonstrating that 3.8B parameters with heavy data filtering and synthetic data could reach MMLU 70%, matching much larger 2023-era models. A 128K-context variant shipped alongside via LongRoPE.
Context · B
Announced at OpenAI DevDay on November 6, 2023 as a 128K-context, cheaper successor to the original GPT-4 endpoint. The gpt-4-turbo-2024-04-09 revision shipped as the general-availability version with vision support and a knowledge cutoff through December 2023.