[Integrations](https://spiral.ai/integrations) › [Instinct](https://spiral.ai/integrations/instinct) › [DeepSeek V3.2 Exp](https://spiral.ai/integrations/instinct/deepseek-v3-2-exp)

# How to Use DeepSeek V3.2 Exp in Instinct

Use DeepSeek V3.2 Exp for coding, math problems, and analysis of long text. Compared with DeepSeek V3.1 Terminus, it introduces sparse attention for long-context efficiency while keeping broadly similar benchmark performance.

|  |  |
| --- | --- |
| Model | DeepSeek V3.2 Exp |
| Available through | OpenRouter |
| Category | LLM |
| Price | Up to $0.002196 per run at the default inputs. $0.351 per 1M input tokens, $0.533 per 1M output tokens. |
| Returns | text, json |
| Runs | In the background |
| Setup | About 2 minutes, with a Spiral API key |

## Prompts to try with DeepSeek V3.2 Exp in Instinct

- Every morning, use DeepSeek V3.2 Exp to summarize the incident updates supplied for the previous day. Return a timestamped timeline, confirmed causes, unresolved hypotheses, and next checks, citing the update behind each statement.
- Each weekday, use DeepSeek V3.2 Exp to classify the support messages supplied for that run. Return JSON following {"tickets":[{"id":"T-101","category":"bug","summary":"Export fails","evidence":"Export stops at 50%"}]}. Preserve message IDs and quote evidence.
- Each week, use DeepSeek V3.2 Exp to compare the current and previous requirements text supplied for that run. List additions, removals, contradictions, and decisions requiring review, with section references.
- Every Friday, use DeepSeek V3.2 Exp to review the research excerpts collected during the week. Produce a 600-word brief separating supported findings, disagreements between sources, and unanswered questions. Cite the supplied source labels.
- After each nightly test run, use DeepSeek V3.2 Exp with temperature set to zero to analyze the supplied failure messages and code excerpts. Group related failures, identify likely causes, and propose checks that could confirm each cause.

## How to set up DeepSeek V3.2 Exp in Instinct

Pick the way that suits you. Each takes about two minutes.

### Use the Spiral skill

1. **Send Instinct this message**: Your agent reads Spiral's skill and sets itself up.

```
set up https://spiral.ai/SKILL.md
```

2. **Sign in or paste a key**: Approve Spiral in the browser, or give the agent a key from the console.

### Use Spiral's MCP server

1. **Create an API key**: In the console, open API keys and create a key for Instinct. Copy it.

2. **Send Instinct this message**: Instinct reads Spiral's skill and asks for your API key.

```text
set up https://spiral.ai/SKILL.md
```

3. **Give Instinct the key**: When Instinct asks for the key, add it where Instinct keeps secrets, never in the chat itself.

### Use the Spiral CLI

1. **Install and sign in**: Installs the Spiral CLI, signs you in and adds its skill to your coding agents.

```bash
npx -y spiral-cli@latest init
```

### Call the Spiral API

1. **Create an API key**: In the console, under API keys.

2. **Run DeepSeek V3.2 Exp**: One request; the result comes back by polling or a webhook.

```bash
curl https://api.spiral.ai/v1/runs \
  -H "Authorization: Bearer $SPIRAL_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $(uuidgen)" \
  -d '{
    "provider": "openrouter",
    "model": "deepseek/deepseek-v3.2-exp",
    "input": {
      "messages": [
        {
          "role": "system"
        }
      ]
    }
  }'
```

Then paste one of the prompts above: Instinct finds DeepSeek V3.2 Exp on Spiral, shows the price and runs it.

Setup guide in the docs: https://spiral.ai/docs/quickstarts/instinct

## DeepSeek V3.2 Exp tools and prices

- [DeepSeek: DeepSeek V3.2 Exp](https://spiral.ai/tools/openrouter/deepseek/deepseek-v3.2-exp) (OpenRouter): Up to $0.002196 per run at the default inputs. $0.351 per 1M input tokens, $0.533 per 1M output tokens.

## Settings you can ask for

| Setting | Allowed values | Default | Description |
| --- | --- | --- | --- |
| tools |  | — |  |
| top_k | 0 to … | — |  |
| top_p | 0 to 1 | — |  |
| reasoning |  | — | Reasoning effort; reasoning tokens count toward max_tokens. |
| max_tokens | 1 to 147456 | — | The most output tokens to generate. |
| temperature | 0 to 2 | — |  |

## What can you do with DeepSeek V3.2 Exp in Instinct?

DeepSeek V3.2 Exp can turn recurring batches of text into incident briefs, requirement comparisons, and structured review queues. Define the evidence to inspect and a fixed output structure so each run produces something you can check.

### Debug a failing test

Give DeepSeek V3.2 Exp the failure output and relevant code. Request a minimal patch, an explanation tied to the failure, and a regression test.

### Review a proposed code change

Ask DeepSeek V3.2 Exp to examine a diff for behavior changes, edge cases, and missing tests. Require each finding to identify the affected code and a reproducible trigger.

### Compare lengthy specifications

Provide the text of two specifications and ask DeepSeek V3.2 Exp to identify changed requirements, contradictions, and unresolved decisions with section references.

### Extract records from text

Ask DeepSeek V3.2 Exp to turn support messages or meeting notes into JSON. Supply an example record and require null for details absent from the source.

### Check a mathematical solution

Give DeepSeek V3.2 Exp a problem and a proposed solution. Ask it to find the first unsupported step, test boundary cases, and state a corrected result.

### Prepare a research comparison

Supply source passages and ask DeepSeek V3.2 Exp to compare their claims. Require supporting excerpts and a separate list of questions the supplied evidence cannot answer.

### Draft an incident handoff

Provide timestamped incident updates and ask DeepSeek V3.2 Exp to assemble a timeline, confirmed findings, competing explanations, and the next diagnostic checks.

## Overview

- [OpenRouter](https://spiral.ai/providers/openrouter)
- [All Instinct guides](https://spiral.ai/integrations/instinct)

## DeepSeek V3.2 Exp in other agents

- [Amp](https://spiral.ai/integrations/amp/deepseek-v3-2-exp)
- [ChatGPT](https://spiral.ai/integrations/chatgpt/deepseek-v3-2-exp)
- [Claude](https://spiral.ai/integrations/claude-desktop/deepseek-v3-2-exp)
- [Claude Code](https://spiral.ai/integrations/claude-code/deepseek-v3-2-exp)
- [Cline](https://spiral.ai/integrations/cline/deepseek-v3-2-exp)
- [Codex](https://spiral.ai/integrations/codex/deepseek-v3-2-exp)
- [Cursor](https://spiral.ai/integrations/cursor/deepseek-v3-2-exp)
- [Devin](https://spiral.ai/integrations/devin/deepseek-v3-2-exp)
- [dots](https://spiral.ai/integrations/dots/deepseek-v3-2-exp)
- [Factory Droid](https://spiral.ai/integrations/factory-droid/deepseek-v3-2-exp)
- [Gemini](https://spiral.ai/integrations/gemini/deepseek-v3-2-exp)
- [Gemini CLI](https://spiral.ai/integrations/gemini-cli/deepseek-v3-2-exp)
- [Google Antigravity](https://spiral.ai/integrations/antigravity/deepseek-v3-2-exp)
- [goose](https://spiral.ai/integrations/goose/deepseek-v3-2-exp)
- [Grok](https://spiral.ai/integrations/grok/deepseek-v3-2-exp)
- [Grok Build](https://spiral.ai/integrations/grok-build/deepseek-v3-2-exp)
- [Hermes](https://spiral.ai/integrations/hermes/deepseek-v3-2-exp)
- [Manus](https://spiral.ai/integrations/manus/deepseek-v3-2-exp)
- [Muse](https://spiral.ai/integrations/muse/deepseek-v3-2-exp)
- [Muse Code](https://spiral.ai/integrations/muse-code/deepseek-v3-2-exp)
- [n8n](https://spiral.ai/integrations/n8n/deepseek-v3-2-exp)
- [Notion](https://spiral.ai/integrations/notion/deepseek-v3-2-exp)
- [OpenClaw](https://spiral.ai/integrations/openclaw/deepseek-v3-2-exp)
- [opencode](https://spiral.ai/integrations/opencode/deepseek-v3-2-exp)
- [Perplexity](https://spiral.ai/integrations/perplexity/deepseek-v3-2-exp)
- [Raycast](https://spiral.ai/integrations/raycast/deepseek-v3-2-exp)
- [VS Code](https://spiral.ai/integrations/vscode/deepseek-v3-2-exp)
- [Windsurf](https://spiral.ai/integrations/windsurf/deepseek-v3-2-exp)
- [Zapier](https://spiral.ai/integrations/zapier/deepseek-v3-2-exp)

## More Instinct guides

- [DeepSeek V3.2](https://spiral.ai/integrations/instinct/deepseek-v3-2)
- [DeepSeek V4 Flash 0423](https://spiral.ai/integrations/instinct/deepseek-v4-flash-0423)
- [DeepSeek V3.1 Terminus](https://spiral.ai/integrations/instinct/deepseek-v3-1-terminus)
- [DeepSeek V4 Flash 0731](https://spiral.ai/integrations/instinct/deepseek-v4-flash-0731)
- [DeepSeek V3.1](https://spiral.ai/integrations/instinct/deepseek-v3-1)
- [DeepSeek V4 Flash 0731 (batch)](https://spiral.ai/integrations/instinct/deepseek-v4-flash-0731-batch)

## Questions about DeepSeek V3.2 Exp in Instinct

**Can Instinct use DeepSeek V3.2 Exp?**

Yes. Add Spiral to Instinct as an MCP server, and Instinct can find and run DeepSeek V3.2 Exp like any built-in tool.

**How much does DeepSeek V3.2 Exp cost in Instinct?**

Up to $0.002196 per run at the default inputs. $0.351 per 1M input tokens, $0.533 per 1M output tokens. You pay from your Spiral balance as you go, with no subscription, and a failed run is not charged. Your Instinct plan is separate.

**Do I need an OpenRouter account to use DeepSeek V3.2 Exp in Instinct?**

You need no OpenRouter account. Instinct uses one Spiral API key for every tool, and Spiral runs DeepSeek V3.2 Exp on its own OpenRouter account.

**How long does it take to set up DeepSeek V3.2 Exp in Instinct?**

About two minutes: create an API key, then send Instinct this message, then give Instinct the key.

**Does Instinct wait while DeepSeek V3.2 Exp runs?**

DeepSeek V3.2 Exp runs in the background. Instinct starts the run, checks on it and gives you the text, json when it's ready.

**What is DeepSeek V3.2 Exp best suited for?**

Coding, math problem solving, and analysis of long text are useful starting points. DeepSeek's published results include 74.1 on LiveCodeBench and 89.3 on AIME 2025; test it on representative work before relying on its answers.

**What changed from DeepSeek V3.1 Terminus?**

DeepSeek V3.2 Exp introduces DeepSeek Sparse Attention to improve efficiency when processing long context. DeepSeek reports broadly similar benchmark performance, so the main change concerns efficiency rather than a large increase in answer accuracy.

**How much text can DeepSeek V3.2 Exp handle?**

Its documented context window is 163,840 tokens. As a working practice, leave room for the answer and send the passages needed for the task before adding background material.

**Where does DeepSeek V3.2 Exp struggle?**

Difficult reasoning remains a limitation: DeepSeek reports 19.8 on Humanity's Last Exam in reasoning mode without tools. Ask for checkable evidence, run generated code, and verify consequential conclusions.

**How should I ask DeepSeek V3.2 Exp for JSON?**

Include the word json and show a small example of the exact structure you want. Specify how to represent missing information, then validate the result before using it.

