How much does prompt caching save on DeepSeek-V3.2?

Prompt caching on DeepSeek-V3.2 bills cached input at $0.0280 per 1,000,000 token instead of $0.2800 per 1,000,000 token for standard input. The saving applies to the cached prefix only; the rest of the request is billed normally.

How does DeepSeek-V3.2 score on benchmarks?

Recorded results for DeepSeek-V3.2: aa_intelligence_index 35.7; arena_text 85.2. Benchmark positions are a coarse filter only — run your own prompts against the candidate models before you commit, because ranking shifts with task shape and prompt style.

What is the rate limit for DeepSeek-V3.2?

The current model record does not publish a numeric rate limit for DeepSeek-V3.2. Confirm the quota for your plan in the console, then run a short load test at your expected concurrency to establish the effective ceiling before production.

How does DeepSeek-V3.2 compare with deepseek-v3-2-251201, and what should you compare?

DeepSeek-V3.2 sits alongside deepseek-v3-2-251201 , qwen3.5-27b , grok-4-0709 in the catalog. Compare them on the dimensions that change your bill and your users’ experience: task quality on your own prompts, latency at your concurrency, context and output limits, and cost per completed task rather than per token.

What is the max output length for DeepSeek-V3.2?

DeepSeek-V3.2 returns up to 65,536 output tokens per request, within a 163,840-token context window. Cap the response with the max_tokens parameter so a single call cannot run past your budget.

Model detail

DeepSeek-V3.2

Provided by DeepSeek

Pay-as-you-go

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 that reduces training and inference cost while preserving quality in long-context scenarios. A scalable reinforcement learning post-training framework further improves reasoning, with reported performance in the GPT-5 class, and the model has demonstrated gold-medal results on the 2025 IMO and IOI. V3.2 also uses a large-scale agentic task synthesis pipeline to better integrate reasoning into tool-use settings, boosting compliance and generalization in interactive environments. Users can control the reasoning behaviour with the `reasoning` `enabled` boolean. [Learn more in our docs](https://openrouter.ai/docs/use-cases/reasoning-tokens#enable-reasoning-with-default-config)

Model specs

Model ID: deepseek-v3.2

Context length: 163.84K
Max output: 65.536K
I/O modalities
Released: 2025-12

API endpoints

One MixRoute gateway, OpenAI-compatible

OpenAI-compatible /v1/chat/completions POST

Pricing

Flat rate, unit: /1M Tokens

Input

$0.2800 /1M Tokens

Completion

$0.4200 /1M Tokens

Cache read

$0.0280 /1M Tokens

Selection summary

Quickly judge whether it fits your workload

What is this model good for?

Large language model designed to harmonize high computational efficiency

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

What should you check before using it?

163,840 context, 65,536 output tokens

Model type: text. Verify supported features and quotas with the provider documentation. Knowledge cutoff: 2024-07. Released 2025-12-01.

Why use it through MixRoute?

DeepSeek-V3.2 runs tools and structured output here

Tool calling and structured output run on that route as well. MixRoute serves DeepSeek-V3.2 on the OpenAI-compatible endpoint /v1/chat/completions; point your existing client at https://api.mixroute.ai/v1 with the model ID deepseek-v3.2 and re-run your integration tests.

Capabilities & compatibility

Unconfirmed capabilities stay "Unknown," no guesses

Tool calling

DeepSeek-V3.2 supports function and tool calls.

EX Send a tools definition; the response returns tool_calls to execute and feed results back.

Structured output

DeepSeek-V3.2 can produce JSON-formatted output on demand.

EX response_format: { type: "json_schema" } forces valid JSON output.

Streaming

DeepSeek-V3.2 supports streamed responses.

EX stream: true streams tokens incrementally, so chat UIs render as they arrive.

Prompt caching

DeepSeek-V3.2 supports prompt caching under supported routing conditions.

EX Keep the system prompt and tool definitions at the start of the request; repeated prefixes are automatically billed at the cache-read rate.

OpenAI SDK

Call DeepSeek-V3.2 through MixRoute's OpenAI-compatible endpoint.

EX Change base_url to https://api.mixroute.ai/v1; everything else follows the OpenAI style.

Token cost estimator

Live estimate from this page's pricing, not an actual bill

Input tokens Max per request 163.84K

Output tokens Max per request 65.536K

Requests / month

Cache hit rate 0%

Share of the same prefix read repeatedly, up to 100%

FAQ

Set max_tokens on DeepSeek-V3.2 to the longest response your task actually needs, not the 65,536-token ceiling. Keeping the cap tight reduces worst-case spend and shortens tail latency. Raise it only for tasks that demonstrably need longer output.

Get started

01

Create an API key

Generate a key in the Console; one key calls every model.
02

Set the base URL

Fully OpenAI SDK-compatible; just swap the URL to connect.

https://api.mixroute.ai/v1
03

Specify the model ID

In the request's model field, enter:

deepseek-v3.2

Create an API key

New users get $5 in free credits

Test this model and alternate routes with the same request format

Start from a real workload, then let quality, total cost, and failure conditions decide whether to send production traffic.

Create an API key View integration docs