DeepSeek-V3.1

DeepSeek-V3.1 is a hybrid model supporting both thinking and non-thinking modes through different chat templates. Built on DeepSeek-V3.1-Base with a two-phase long context extension (32K phase: 630B tokens, 128K phase: 209B tokens), it features 671B total parameters with 37B activated. Key improvements include smarter tool calling through post-training optimization, higher thinking efficiency achieving comparable quality to DeepSeek-R1-0528 while responding more quickly, and UE8M0 FP8 scale data format for model weights and activations. The model excels in both reasoning tasks (thinking mode) and practical applications (non-thinking mode), with particularly strong performance in code agent tasks, math competitions, and search-based problem solving.

Benchmark results

Benchmark Score Tags Source
Aider-Polyglot 68.4% self-reported llm-stats link →
AIME 2024 66.3% self-reported llm-stats link →
AIME 2025 49.8% self-reported llm-stats link →
BrowseComp 30.0% self-reported llm-stats link →
BrowseComp-zh 49.2% self-reported llm-stats link →
CodeForces 69.7% self-reported llm-stats link →
GPQA 74.9% self-reported llm-stats link →
HMMT 2025 33.5% self-reported llm-stats link →
Humanity's Last Exam 15.9% self-reported llm-stats link →
LiveCodeBench 56.4% self-reported llm-stats link →
MMLU-Pro 83.7% self-reported llm-stats link →
MMLU-Redux 91.8% self-reported llm-stats link →
SimpleQA 93.4% self-reported llm-stats link →
SWE-bench Multilingual 54.5% self-reported llm-stats link →
SWE-Bench Verified 66.0% self-reported llm-stats link →
Terminal-Bench 31.3% self-reported llm-stats link →