impulse

DeepSeek V3.2

Open weights

DeepSeek-V3.2 is a large language model designed to harmonize high computational efficiency with strong reasoning and agentic tool-use performance.

API id deepseek-v3.2Released 1 Dec 2025 (Dec 2025)
Input / 1M
$0.28
at DeepSeek
Output / 1M
$0.40
at DeepSeek
Context window
164K
tokens
Max output
164K
tokens per reply
Hosts
8
8 listings
Released
Dec 2025
1 Dec 2025

Pricing

List prices at DeepSeek, in US dollars per million tokens.

TierInput / 1MOutput / 1M
Standard$0.28$0.40
Cached input
90% off input
$0.028

What it costs a month

At DeepSeek list prices, 30 days.

  • Chat assistant5,000 requests a day, 2K in / 0.5K out, cached$91.32
  • RAG over documents2,000 requests a day, 8K in / 0.6K out, cached$88.32
  • Coding agent1,000 requests a day, 40K in / 2K out, cached$148
  • Overnight batch20,000 requests a day, 4K in / 0.4K out$768
Use your own numbers

Where to run it

8 hosts serve DeepSeek V3.2. The comparison is on blended price, 3 parts input to 1 part output.

HostIn / out per 1M
DeepSeekLab API
deepseek-v3.2
$0.28 / $0.40
DeepInfraInference
DeepSeek-V3.2
$0.26 / $0.38
−6% vs headline
Novita AIInference
deepseek-v3.2
$0.269 / $0.40
−3% vs headline
GMI CloudInference
DeepSeek-V3.2
$0.28 / $0.40
deepseek-v3.2-maas
$0.56 / $1.68
+171% vs headline
Fireworks AIInference
deepseek-v3p2
$0.56 / $1.68
+171% vs headline
deepseek-v3.2
$0.58 / $1.68
+176% vs headline
Retires 1 Nov 2027
SambaNovaInference
DeepSeek-V3.2
$3.00 / $4.50
+989% vs headline

Capabilities

Core

  • Streaming
  • Function calling
  • JSON mode
  • Structured outputs

Multimodal

  • Vision
  • Image input
  • Audio
  • Image generation
  • Video

Advanced

  • Reasoning
  • Tool calling
  • MCP
  • Prompt caching
  • Batch API
  • Embeddings
Takes in
Text
Produces
Text
Features
Assistant prefillPrompt caching
Open weights
License: DeepSeek License (MIT code)
Runs in about 1644.9 GB of GPU memory
Runtimes: SGLang, TGI, vLLM

Cheaper alternatives

Current models that cost less and keep most of what this one can do: tools, vision, reasoning, structured output, caching and enough context.

Same model, cheaper host

  • DeepInfra
    DeepSeek-V3.2
    $0.26 / $0.38
    6% less
  • Novita AI
    deepseek-v3.2
    $0.269 / $0.40
    3% less

Smaller DeepSeek models

No cheaper DeepSeek model keeps its key capabilities.

Other frontier labs

  • Claude Haiku 5.5
    Anthropic · on Anthropic
    $0.10 / $0.50
    Has every key capability
    35% less
  • GPT-6 Luna
    OpenAI · on OpenAI
    $0.10 / $0.50
    Has every key capability
    35% less
  • GPT-4o Mini
    OpenAI · on OpenAI
    $0.15 / $0.60
    Lacks reasoning
    15% less

Open-weight picks

  • Deepseekv4.1 Flash
    DeepSeek · on Sarvam
    $0.149 / $0.585
    Has every key capability
    17% less
  • Qwen3.8 27B
    Alibaba Qwen · on EmpirioLabs AI
    $0.17 / $0.50
    Has every key capability
    19% less
  • Nemotron 3.5 Super VL
    NVIDIA · on Kyma API
    $0.081 / $0.472
    Has every key capability
    42% less
  • MiMo V2.6 Flash
    Xiaomi · on GMI Cloud
    $0.112 / $0.224
    Has every key capability
    55% less

Model line

Newer and older versions, at each lab's own price.

  1. UndatedDeepSeek V3.2.251201LatestNot listed
  2. Dec 2025DeepSeek V3.2This page$0.28 / $0.40
  3. UndatedDeepSeek V3.1$0.25 / $0.95
  4. UndatedDeepSeek V3$0.89 / $0.89
  5. UndatedDeepSeek V2.5$1.20 / $1.20

Call it

Using DeepSeek's API.

Python
from litellm import completion

response = completion(
    model="deepseek/deepseek-v3.2",
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

Quick answers

How much does DeepSeek V3.2 cost?

On DeepSeek, DeepSeek V3.2 costs $0.28 per million input tokens and $0.40 per million output tokens. Cached input costs $0.028 per million tokens. The cheapest host is DeepInfra, at $0.26 per million input tokens and $0.38 per million output tokens.

Which APIs offer DeepSeek V3.2?

DeepSeek V3.2 runs on 8 APIs: DeepSeek, DeepInfra, Novita AI, GMI Cloud, Vertex AI, Fireworks AI, Azure AI Foundry and SambaNova.

What is DeepSeek V3.2's context window?

164K tokens, with up to 164K tokens in one reply.

Is DeepSeek V3.2 being retired?

Azure AI Foundry retires it on 1 Nov 2027. 7 hosts keep serving it.

Is there a newer version of DeepSeek V3.2?

Yes. DeepSeek V3.2.251201 is the newest model in this line.

What are cheaper alternatives to DeepSeek V3.2?

Models that cost less and keep most of what it can do: Claude Haiku 5.5 on Anthropic (35% less), GPT-6 Luna on OpenAI (35% less) and Deepseekv4.1 Flash on Sarvam (17% less). DeepSeek V3.2 itself costs 6% less on DeepInfra. Savings are on a mix of 3 input tokens to 1 output token.

When was DeepSeek V3.2 released?

DeepSeek released it on 1 Dec 2025.

What is DeepSeek V3.2's API model id?

Use deepseek-v3.2 with DeepSeek. Other hosts use their own ids: DeepSeek-V3.2 on DeepInfra, DeepSeek-V3.2 on GMI Cloud, deepseek-v3.2-maas on Vertex AI, deepseek-v3p2 on Fireworks AI.

Sources

Raw catalog entry for deepseek-v3.2 on DeepSeek
{
  "cache_read_input_token_cost": 2.8e-8,
  "input_cost_per_token": 2.8e-7,
  "input_cost_per_token_cache_hit": 2.8e-8,
  "litellm_provider": "deepseek",
  "max_input_tokens": 163840,
  "max_output_tokens": 163840,
  "mode": "chat",
  "output_cost_per_token": 4e-7,
  "supports_assistant_prefill": true,
  "supports_function_calling": true,
  "supports_prompt_caching": true,
  "supports_reasoning": true,
  "supports_tool_choice": true
}