DeepSeek V3.2 pricing & benchmarks
DeepSeek V3.2 is available from DeepSeek at $0.260 per million input tokens and $0.380 per million output tokens ($0.290 blended at 3:1). It accepts up to 163,840 tokens of context. 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...
Pricing
Per million tokens · OpenRouter
Input$0.260
Output$0.380
Blended (3:1)$0.290
Specification
deepseek/deepseek-v3.2
Context window164K
Max output164K
ReleasedDec 2025
LicenceOpen weights
Modalitiesin:text, out:text
Benchmarks
Independently run — not vendor self-reports. Every score links its source.
| Benchmark | Score | Config | Run | Source |
|---|---|---|---|---|
| GPQA Diamond | 71.2% ±3.2 | — | 2026-07-16 | Independently run · Epoch AI |
| OTIS Mock AIME | 48.9% ±7.5 | — | 2026-07-16 | Independently run · Epoch AI |
Scores published by Epoch AI, 'AI Benchmarking Hub', used under CC BY 4.0.
Cheaper models that score at least as well on GPQA Diamond
Strictly lower blended price AND an equal-or-higher GPQA Diamond score. One benchmark is one dimension — a model that wins here may still be weaker at your specific task.
| Model | Provider | Blended $/1M | GPQA Diamond | |
|---|---|---|---|---|
| gpt-oss-120b | OpenAI | $0.070 (−76%) | 75.8% | Compare → |
| DeepSeek V4 Flash 0731 | DeepSeek | $0.113 (−61%) | 91.0% | Compare → |
| Gemma 4 31B | $0.160 (−45%) | 75.8% | Compare → | |
| GPT-5.6 Luna | OpenAI | $0.225 (−22%) | 91.6% | Compare → |
Where to run DeepSeek V3.2
14 hosts · 14× between cheapest and dearest. Blended at 3:1. ⚠ marks an option serving less context than the best available, or re-pricing above a prompt-length threshold. Bracketed labels are the provider's own service tier or region.
| Host | Blended $/1M | In / Out | Context | Weights | Uptime 24h |
|---|---|---|---|---|---|
| Baidu ⚠ 131K context, not 164K |
$0.233 | $0.207 / $0.311 | 131K | fp8 | 99.8% |
| StreamLake ⚠ 128K context, not 164K |
$0.241 | $0.214 / $0.322 | 128K | fp8 | 99.9% |
| DeepInfra | $0.290 | $0.260 / $0.380 | 164K | fp4 | 98.0% |
| AtlasCloud | $0.290 | $0.260 / $0.380 | 164K | fp8 | 99.9% |
| SiliconFlow | $0.299 | $0.259 / $0.420 | 164K | fp8 | 99.8% |
| Novita | $0.302 | $0.269 / $0.400 | 164K | fp8 | 99.9% |
| GMICloud | $0.325 | $0.290 / $0.430 | 164K | fp8 | 99.8% |
| Venice | $0.367 | $0.330 / $0.480 | 160K | — | 93.8% |
| DigitalOcean | $0.387 | $0.250 / $0.800 | 164K | — | 98.0% |
| Alibaba ⚠ 131K context, not 164K |
$0.556 | $0.370 / $1.11 | 131K | fp8 | 98.7% |
| Friendli | $0.750 | $0.500 / $1.50 | 164K | — | 99.9% |
| $0.840 | $0.560 / $1.68 | 164K | — | 99.8% | |
| Phala | $1.00 | $1.00 / $1.00 | 164K | — | 98.2% |
| SambaNova ⚠ 33K context, not 164K |
$3.38 | $3.00 / $4.50 | 33K | — | 95.5% |
Host prices and uptime from OpenRouter, synced 2026-08-08. Lower-precision weights (fp4, fp8) can cost less and score worse than the same model at bf16 — the score above is not host-specific.
Compare DeepSeek V3.2
FAQ
How much does DeepSeek V3.2 cost?
DeepSeek V3.2 costs $0.260 per million input tokens and $0.380 per million output tokens on OpenRouter. At a 3:1 input:output mix that blends to $0.290 per million tokens.
What is DeepSeek V3.2's context window?
DeepSeek V3.2 accepts up to 163,840 tokens of context, and can generate up to 163,840 output tokens.
How good is DeepSeek V3.2 on benchmarks?
DeepSeek V3.2 scores 71.2% on GPQA Diamond, independently run and published by Epoch AI. 2 benchmark results are listed on this page, each with its run date and source.
Is there a cheaper model as good as DeepSeek V3.2?
Yes — 4 models in our catalogue cost less per token than DeepSeek V3.2 and score at least as high on the same benchmark. The cheapest is gpt-oss-120b at $0.070 per million tokens blended.