DeepSeek V3.2 Speciale
DeepSeek V3.2 Speciale — maxed-out reasoning capabilities rivaling Gemini-3.0-Pro. Gold-medal performance on IMO, CMO, and ICPC World Finals. 65.5K max output with DeepSeek Sparse Attention. FP8.
- Reasoning
- Native PDF input
Added Dec 2, 2025
Pricing
Auto routing · per 1M tokens- Input
- $0.28
- Output
- $0.42
Specifications
- Context window
- 163K
- Max output
- 65.5K
- Parameters
- 685B / 37B
- Total / active
Benchmarks
Benchmarks
Sourced from Artificial Analysis.
Intelligence Index
14.5
Agentic work
T²-Bench Telecom (legacy)
Legacy fallback · Conversational AI agents in dual-control scenarios
0.0%
Better than 3% of models compared
Document reasoning
AA-LCR v1.1
Long context reasoning with updated grading
70.0%
Better than 61% of models compared
Reasoning
HLE
Humanity's Last Exam
28.7%
Better than 75% of models compared
IFBench
Instruction-following benchmark
63.9%
Better than 75% of models compared
Coding
Terminal-Bench Hard (legacy)
Legacy fallback · Agentic coding and terminal use
34.8%
Better than 80% of models compared
LiveCodeBench
Contamination-free coding benchmark
89.6%
Better than 99% of models compared
Math
AIME 2025
American Invitational Mathematics Examination 2025
96.7%
Better than 98% of models compared
Knowledge
MMLU-Pro
Professional and academic subject knowledge
86.3%
Better than 95% of models compared
Legacy benchmarks
GPQA Diamond (legacy)
Graduate-level scientific reasoning
87.1%
Better than 83% of models compared
AA-LCR (unversioned / legacy)
Long context reasoning evaluation
70.0%
Better than 61% of models compared
Last updated Oct 19, 2026
Artificial Analysis