Early access: features, availability, and pricing may change as Kyara Intelligence evolves.

Below baseline Reasoning Obsolete

DeepSeek: DeepSeek V3.2

Experimental architecture with sparse attention for efficient long-context RP and reasoning tasks.

This is a legacy model, superseded by newer releases. It remains fully available to use.

Model ID deepseek/deepseek-v3.2

Quick start

Drop this model into any OpenAI-compatible client by setting the base URL, your API key, and the model ID below.

curl https://api.kyara-intelligence.com/v1/chat/completions \
  -H "Authorization: Bearer $KYARA_INTELLIGENCE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v3.2",
    "messages": [
      { "role": "user", "content": "Hello!" }
    ]
  }'

See the full API reference

Overview

DeepSeek Experimental is an experimental large language model that introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism designed to improve training and inference efficiency in long-context scenarios. With a massive 163,840 token context window, it performs on par with the standard DeepSeek model across reasoning, coding, and agentic tool-use tasks. This research-oriented model focuses on validating architectural optimizations for extended context lengths while maintaining quality.

Specifications

Context / Memory 📚 Standard · 160K tokens

Cost vs baseline Below baseline

Input

0.5x

Output

0.3x

Input

4.1M

cr / M tokens

Output

6.2M

cr / M tokens

Multiples are relative to the catalog median; bars are scaled to the most expensive model.

Model ID

deepseek/deepseek-v3.2

Reasoning

Available

Status

Obsolete

Upstream moderation

All models on Kyara Intelligence are accessed through their original API providers (Mistral, Z.AI/GLM, xAI, DeepSeek, and others). Each provider's content policies and moderation apply to all requests.

Kyara routes API calls without modifying inputs or outputs, and does not log prompt or response content. Users are responsible for complying with the relevant provider's terms of service in addition to Kyara's terms.