The Open-Source AI Stack
RSS
All models

Models · Compare

DeepSeek-V3.1 vs GPT-5

Rows highlighted in warm gray are where the models differ. Numbers carry their as-of date and primary source.

Specs

Field A: DeepSeek-V3.1 B: GPT-5
Released
Developer DeepSeekOpenAI
Openness OpenProprietary
License MITProprietary
OSI-approved yesno
Data released nono
Training code nono
Architecture moeunknown
Total params 671B
Active params 37B
Experts
Context window 128K
Attention mlaunknown
Position enc. rope-yarnunknown
Pretraining tokens
Post-training sft, rlhfrlhf
Training hardware
$/M input $0.56$1.25
$/M output $1.67$10.00
Output tok/sec 072

Benchmarks

Missing scores render as not reported; never inferred. Bold highlights the leader per benchmark.

General reasoning

MMLU-Pro 83.3 2026-05-21 87.1 2026-05-21
GPQA-Diamond 73.5 2026-05-21 85.4 2026-05-21

Code

LiveCodeBench 57.7 2026-05-21 84.6 2026-05-21

Math

MATH 99.4 2026-05-21
AIME 2024 95.7 2026-05-21
AIME 2025 49.7 2026-05-21

Context · A

Hybrid V3.1 with one model serving both thinking and non-thinking modes, released August 21 2025. DeepSeek added 840B tokens of continued pretraining for long-context extension (32K and 128K phases) and shipped under MIT. Pricing update took effect September 5 2025.

Context · B

OpenAI's first unified model, fusing the fast GPT-series with the deeper o-series reasoning track behind a real-time router that picks per-turn. Launched August 7 2025 via livestream; immediately default in ChatGPT, available in API and on the GitHub Models Playground. Ships in main, mini, nano, thinking, and thinking-mini variants with low/medium/high/minimal effort settings.

DeepSeek-V3.1 detail → · GPT-5 detail →