Models · Compare
DeepSeek-V2 Chat vs GPT-4o
Rows highlighted in warm gray are where the models differ. Numbers carry their as-of date and primary source.
Specs
| Field | A: DeepSeek-V2 Chat | B: GPT-4o |
|---|---|---|
| Released | 2024-05-07 | 2024-05-13 |
| Developer | DeepSeek | OpenAI |
| Openness | Open | Proprietary |
| License | DeepSeek License | Proprietary |
| OSI-approved | no | no |
| Data released | no | no |
| Training code | no | no |
| Architecture | moe | unknown |
| Total params | 236B | — |
| Active params | 21B | — |
| Experts | — | — |
| Context window | 128K | — |
| Attention | mla | unknown |
| Position enc. | rope-yarn | unknown |
| Pretraining tokens | 8.1T | — |
| Post-training | sft, rlhf | rlhf |
| Training hardware | — | — |
| $/M input | — | $2.50 |
| $/M output | — | $10.00 |
| Output tok/sec | — | 131.6 |
Benchmarks
Missing scores render as not reported; never inferred. Bold highlights the leader per benchmark.
General reasoning
| MMLU-Pro | — | 74.8 2026-05-21 |
| GPQA-Diamond | — | 54.3 2026-05-21 |
Code
| LiveCodeBench | — | 30.9 2026-05-21 |
Math
| MATH | — | 75.9 2026-05-21 |
| AIME 2024 | — | 15.0 2026-05-21 |
| AIME 2025 | — | 6.0 2026-05-21 |
Context · A
The Multi-head Latent Attention (MLA) debut paper. Cut KV-cache memory by ~93% versus dense attention, making 128K-context MoE inference economically viable.
Context · B
Native-multimodal model with audio + vision + text in a single pretrained backbone. Pushed real-time voice latency to under 400ms; the multimodal benchmark anchor through 2024.