Models · Compare
DeepSeek-V4 Pro vs GPT-5.5
Rows highlighted in warm gray are where the models differ. Numbers carry their as-of date and primary source.
Specs
| Field | A: DeepSeek-V4 Pro | B: GPT-5.5 |
|---|---|---|
| Released | — | 2026-04-23 |
| Developer | DeepSeek | OpenAI |
| Openness | Open | Proprietary |
| License | MIT | Proprietary |
| OSI-approved | yes | no |
| Data released | no | no |
| Training code | no | no |
| Architecture | moe | unknown |
| Total params | 1.6T | — |
| Active params | 49B | — |
| Experts | — | — |
| Context window | 1.0M | 1.0M |
| Attention | hybrid-csa-hca | unknown |
| Position enc. | unknown | unknown |
| Pretraining tokens | 32.0T | — |
| Post-training | sft, grpo, distillation | rlhf |
| Training hardware | — | — |
| $/M input | $1.74 | $5.00 |
| $/M output | $3.48 | $30.00 |
| Output tok/sec | 29.8 | 66.7 |
Benchmarks
Missing scores render as not reported; never inferred. Bold highlights the leader per benchmark.
General reasoning
| MMLU-Pro | 87.5 2026-04-24 | — |
| GPQA-Diamond | 88.8 2026-05-21 | 93.5 2026-05-21 |
Code
| HumanEval | 76.8 2026-04-24 | — |
| SWE-Bench Verified | 80.6 2026-04-24 | — |
| LiveCodeBench | 93.5 2026-04-24 | — |
Context · A
First V4-generation flagship, released April 24 2026 with a 1.6T-parameter MoE activating 49B per token. Combines Compressed Sparse Attention and Heavily Compressed Attention to cut single-token inference FLOPs to 27 percent of V3.2 and KV cache to 10 percent. 1M context is now the DeepSeek default. V4-Flash (284B / 13B active) shipped alongside.
Context · B
Released April 23 2026, with GPT-5.5 and GPT-5.5 Pro hitting the API the next day. GPT-5.5 Instant replaced GPT-5.3 Instant as the ChatGPT default May 5 2026. Expanded context to 1M tokens; pricing above 272K input doubles input and 1.5x's output for the session.