DeepSeek

DeepSeek offers powerful models at competitive prices, with V3.2 powering both chat and reasoning modes.

Setup

1. Get an API Key

Go to platform.deepseek.com
Sign in or create an account
Navigate to API Keys
Create a new key

2. Configure in BraceKit

Open Settings → AI Provider
Click DeepSeek in the provider grid
Paste your API key
Select a model

Settings are saved automatically as you type.

Available Models

Both models are powered by DeepSeek-V3.2 with different modes:

Model	Mode	Best For	Context	Max Output
deepseek-chat	Non-thinking	General chat, code, summarization	128K	8K tokens
deepseek-reasoner	Thinking	Math, logic, complex analysis	128K	64K tokens

Features

Reasoning (Thinking Mode)

The deepseek-reasoner model shows its Chain-of-Thought reasoning process:

┌─────────────────────────────────────┐
│ 🧠 Thinking...                    ▾ │
├─────────────────────────────────────┤
│ Let me analyze this problem...      │
│                                     │
│ Step 1: Identify the key variables  │
│ Step 2: Consider edge cases         │
│ Step 3: Formulate solution          │
└─────────────────────────────────────┘

Based on my analysis...

This happens automatically with the reasoner model.

Function Calling

DeepSeek supports tool calling for:

MCP server tools
Built-in tools (Google Search)

Cost-Effective

DeepSeek offers very competitive pricing while maintaining high quality.

Model Parameters

Configure in Settings → AI Provider under the Advanced section:

Parameter	Range	Effect
Temperature	0-2	Higher = more creative
Max Tokens	1-8K (chat) / 1-64K (reasoner)	Maximum response length

Recommended Settings

Use Case	Model	Temperature
Code generation	deepseek-chat	0.3
General chat	deepseek-chat	0.7
Complex reasoning	deepseek-reasoner	0.5
Math/Logic	deepseek-reasoner	0.0

Pricing

DeepSeek V3.2 offers unified pricing for both models with automatic context caching:

Type	Price (per 1M tokens)
Input (Cache Hit)	$0.028
Input (Cache Miss)	$0.28
Output	$0.42

Cache Benefits:

Automatic context caching (enabled by default)
90% discount on cached input tokens
Shared prefix across requests triggers caching

Note: Check DeepSeek pricing for current rates.

Troubleshooting

“Rate limit exceeded”

Wait a moment and retry
Check your usage limits in the console

Reasoning not showing

Ensure you’re using deepseek-reasoner (not deepseek-chat)
The thinking mode is optimized for complex queries (math, logic, code)
Simple queries may not trigger extended Chain-of-Thought

Slow responses

The reasoner model takes longer to “think”
For faster responses, use deepseek-chat

OpenAI — Alternative with reasoning models
Anthropic — Alternative with extended thinking
Ollama — Free local alternative