DeepSeek

DeepSeek offers powerful models at competitive prices, with DeepSeek-V4 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
Select DeepSeek from the provider dropdown
Paste your API key
Select a model

Settings are saved automatically as you type.

Available Models

DeepSeek’s current lineup is powered by DeepSeek-V4:

Model	Mode	Best For	Context	Max Output
deepseek-v4-flash	Thinking / Non-thinking	General chat, code, high-throughput tasks	1M	384K tokens
deepseek-v4-pro	Thinking / Non-thinking	Complex reasoning, agentic coding, analysis	1M	384K tokens

Note: the legacy deepseek-chat and deepseek-reasoner names were retired on July 24, 2026 — they now route to deepseek-v4-flash.

Features

Reasoning (Thinking Mode)

The deepseek-v4-pro model shows its Chain-of-Thought reasoning process when thinking mode is enabled:

┌─────────────────────────────────────┐
│ 🧠 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-v4-flash	0.3
General chat	deepseek-v4-flash	0.7
Complex reasoning	deepseek-v4-pro	0.5
Math/Logic	deepseek-v4-pro	0.0

Pricing

DeepSeek V4 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-v4-pro (not the flash variant) for complex reasoning
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-v4-flash

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