DeepSeek V3.1 Terminus

deepseek-ai · released 2025-09-22 · mit license

DeepSeek-V3.1-Terminus (Sept 2025) is a maintenance update of V3.1 improving agent and language consistency, with vendor reasoning-mode scores of 85.0% MMLU-Pro, 80.7% GPQA Diamond and 68.4% SWE-bench Verified.

Key specs

TypeLocal open-weight
Parameters684.53B total · MoE, — active
Architecturedeepseek_v3
Context window164K tokens
Knowledge cutoff2025-03-31
Modalitiestext
Recommended backends
Minimum viable rig

Benchmark scores

GPQA Diamond80.7%
SWE-bench Verified68.4%
AIME
MMLU-Pro85%
BFCL v3 (tool use)
Composite score6.42
Community ratingNo reviews yet

VRAM & disk per quantization

QuantVRAMDiskRAMContext
Q4_K_M398.5 GB397 GB164K

Strengths & weaknesses

Strengths: Hybrid reasoning model with strong agentic tool use (SWE 68.4); Improved language consistency over V3.1; Open weights (MIT), cheap API

Weaknesses: Lower HLE (21.7) than frontier reasoners; AA index of 21 is an estimate