About
DeepSeek-V3.2 is a large language model designed to harmonize high computational efficiency with strong reasoning and agentic tool-use performance. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism...
Related Models
DeepSeek V4 Pro (Reasoning, High Effort)2026-04-24DeepSeek V4 Flash (Reasoning, High Effort)2026-04-24DeepSeek V4 Pro (Non-reasoning)2026-04-24DeepSeek V4 Flash (Non-reasoning)2026-04-24DeepSeek: DeepSeek V4 Flash (free)2026-04-24DeepSeek V4 Pro2026-04-22DeepSeek V4 Flash2026-04-22DeepSeek V3.2 (Reasoning)2025-12-01
Pricing
Input
$0.23
per 1M tokens
Output
$0.34
per 1M tokens
Blended
$0.26
per 1M tokens
Cheaper than 66% of models. Median price is $0.54/1M tokens.
Cost Calculator
Tokens per day1M
100K100M
Daily
$0.26
Monthly
$7.72
vs. Similar Models
MiMo-V2-Flash (Non-reasoning)Q:0.0
$0.15-42%
Gemma 4 31B (Non-reasoning)Q:+0.1
$0.20-20%
Grok 4.3 (Non-reasoning)Q:+0.1
$1.56+507%
Arcee AI: Trinity Large ThinkingQ:-0.2
$0.39+51%
Performance
80
tokens/sec
Faster than 40% of models
0.74
seconds
Faster than 68% of models
0.74
seconds
Faster than 77% of models
Market Median
94 tok/s
15% slower
Median TTFT
1.11s
34% faster
Throughput/Dollar
312
tok/s per $/1M
Speed Comparison
DeepSeek V3.2 Exp (Non-reasoning)
80 tok/s+0%
Magistral Small 1.2
81 tok/s+0%
DeepSeek V3.2 Exp (Reasoning)
80 tok/s-1%
Context Window
131K
tokens
Larger than 27% of models
Max Output
64K
tokens
49% of context
Benchmarks
MMLU-Pro
83.7%
GPQA Diamond
75.1%
HLE
10.5%
LiveCodeBench
59.3%
SciCode
38.7%
TerminalBench Hard
32.6%
MATH-500Not evaluated
AIMENot evaluated
AIME 2025
59.0%
IFBench
49.0%
Long Context Recall
39.0%
Tau2
78.9%
Market AverageTop Score