DeepSeek: DeepSeek V3.2, DeepSeek — route, pricing & source
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...
OpenRoutermediumLast checked: May 14, 2026
Input
$0.2520 / 1M
/ 1M tokens
Output
$0.3780 / 1M
/ 1M tokens
Cached input
Not available
Not available
Cache write
Not available
Not available
Context and limits
Context window
131.1K
Max output
65,536
Input modalities
text
Output modalities
text
Weight access
Closed
Availability
Unknown
Capability flags
ToolsJSONReasoningLong context
From source metadata + verification fields — confirm provider docs before production.
Benchmarks
- LMArena Text Arenaoverall · official_benchmark_api1,425Independent
Rank: 70 | Votes: 9076 | 95% CI: 1418.0-1431.1 | Published: 2026-05-07 | Model license label in dataset: MIT
- LMArena WebDev Arenaoverall · official_benchmark_api1,368Independent
Rank: 46 | Votes: 7917 | 95% CI: 1360.3-1376.1 | Published: 2026-05-07 | LMArena dataset entry: "deepseek-v3.2-thinking" (matched via variant normalization) | Model license label in dataset: MIT
Access and license
Official API
No
Open weights
No
Self-hostable
No
License
Open sourceMITCode/weights ship under an open license (often MIT / Apache‑2.0). MIT is a short permissive license: use/modify/redistribute freely with a copyright notice—always read the upstream project text.
Example costs
Scenario: 1K input + 300 output tokens / request
10,000 requests
$3.65
100,000 requests
$36.54
1,000,000 requests
$365.40
Cheaper alternatives
Lower-cost models for the same scenario
Source
Source
OpenRouter Models API
Confidence
medium
Route
OpenRouter
Last checked
May 14, 2026
Source notes
- Includes Hugging Face benchmark data only when the dataset is public, ungated, allowlisted, and matched by reviewed aliases.
