Model Selection and Provider Configuration

MBAgent itself is not model-locked — you can use Mbbrowser official models, third-party models such as OpenAI / Claude / Gemini / DeepSeek, or your custom gateway or local model. This article details how to choose and configure models.

---

1. What Is a Model?

A model is the AI that actually "thinks" behind MBAgent. MBAgent is the shell; install any brain, and it thinks that way.

Different models have very different capabilities:

Model	Strengths	Weaknesses	Price
deepseek-v4-flash	Fast response, daily chat, ultra-low price	Complex reasoning	⭐ Ultra-cheap
deepseek-v4-pro	Deep reasoning, Chinese understanding	Weak multimodal	⭐⭐ Cheap
gpt-5.4	General chat, fast Q&A	Complex reasoning	⭐ Cheap
gpt-5.5	Comprehensive reasoning, code, general	Higher price	⭐⭐ Mid
gpt-5.6-sol	Complex reasoning, scientific computing	Slower response	⭐⭐⭐ Mid-high
gpt-5.6-terra	Long context, multi-step tasks	High price	⭐⭐⭐ Mid-high
gpt-5.6-luna	Ultra-long context, deep analysis	Low price	⭐⭐ Cheap
claude-sonnet-5	Long-form writing, code, medium tasks	Higher price	⭐⭐⭐ Expensive
claude-opus-5.5	Complex tasks, deep reasoning, long-doc analysis	High price	⭐⭐⭐ Expensive
claude-fabel-5.1	Multimodal, vision, image-text combination	Weaker Chinese	⭐⭐⭐ Expensive

TIP

Beginner recommendation: deepseek-v4-flash or gpt-5.4 — fast, cheap, covers 80% of daily tasks. Switch to a stronger model once familiar.

---

2. What Is a Provider?

A Provider is the AI model vendor. The same model (e.g. gpt-5) may be provided by multiple Providers:

gpt-5 ──┬── OpenAI (official)
        ├── Mbbrowser official (relay, price advantage)
        ├── Azure AI Foundry
        └── Your custom gateway

MBAgent has multiple built-in Providers; each has different model lists, prices, and stability.

NOTE

Settings entry: Click the ⚙️ "Settings" button at the top right of the Chat tab → go to "Provider Configuration" → manage all Providers here. Mbbrowser official models are auto-recognized; other Providers require manual API Key entry.

Screenshot note: Complete Provider configuration popup and vendor selection panel screenshots will be progressively supplemented in v1.x; the current doc is function-led; please follow the text instructions.

Built-in Providers

Provider	Models Served	Price Advantage	Use Case
Mbbrowser official	Full series (incl. deepseek-v4-flash, deepseek-v4-pro, gpt-5.4, gpt-5.5, gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, claude-sonnet-5, claude-opus-5.5, claude-fabel-5.1)	⭐ Cheapest, links with Mbbrowser membership package	First choice for all users
OpenAI official	Full OpenAI series	⭐ High price	Need native vendor support
Anthropic official	Full Claude series	⭐⭐ Mid	Claude power users
Google Gemini official	Gemini / Gemma	⭐⭐ Mid	Multimodal tasks
DeepSeek official	Full DeepSeek series	⭐ Cheap	Budget-sensitive scenarios
DashScope (Alibaba)	Full Qwen series	⭐⭐ Mid	Chinese scenarios
Custom OpenAI-compatible gateway	Custom	Custom	Enterprise users

---

3. Mbbrowser Official Model: The Most Recommended Choice

Mbbrowser official is MBAgent's default Provider, with the following advantages for Mbbrowser users:

1. Bundled with Mbbrowser Membership Package

Mbbrowser membership packages (Free member / Trial member / Basic member / Personal Custom / Personal Unlimited / Team) include AI model quota. Using Mbbrowser official models deducts directly from the package quota.

TIP

View your current quota:

1. Click the "Account" button at the top right of the Chat tab
2. Select "My AI Quota" from the dropdown
3. The Mbbrowser Console auto-opens, showing remaining quota

2. One-Click Access to Mainstream Models

Mbbrowser official has already integrated deepseek-v4-flash, deepseek-v4-pro, gpt-5.4, gpt-5.5, gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, claude-sonnet-5, claude-opus-5.5, claude-fabel-5.1 and other mainstream models. You don't need to apply for an API Key from each vendor yourself; just pick them in MBAgent.

3. Top-up and Invoice

Mbbrowser official models support centralized top-up, expense viewing, and invoicing through the Mbbrowser Console.

---

4. How to Switch Models

Method 1: Model Selector at the Top of the Chat Tab

NOTE

Screenshot note: Clicking the model selector at the top of the Chat tab will pop up the Mbbrowser official model dropdown list; screenshots will be supplemented in v1.x. Please follow the text steps below.

1. Open the Chat tab
2. Click the Model selector at the top (where the current model name is displayed)
3. Pick the model you want from the popup list
4. Takes effect immediately; next input uses the new model

Method 2: Compare Before Picking

If you are torn between multiple models, use the comparison feature:

1. Open the model selector
2. Select 2-4 candidate models
3. Click the "Compare" button
4. See each model's context length / price / speed / capability tags
5. Decide and pick one

NOTE

Screenshot note: Detailed screenshots of the comparison panel (multi-model horizontal comparison view, each model's price curves, capability radar charts, etc.) will be supplemented in v1.x.

Method 3: Set as Default

If you always use the same model, set it as default:

1. Click the "Settings" ⚙️ at the top right
2. Go to "Default Model" option
3. Pick your commonly used model
4. New sessions automatically use this model

---

5. Add a Custom Provider

If you use a third-party LLM relay or a custom gateway, you can add a custom Provider.

Step 1: Get API Info

You need:

• API Key: The key the vendor gives you (typically generated in the vendor's console)
• Base URL: The endpoint address the vendor provides (e.g. https://api.example.com/v1)

Step 2: Configure in MBAgent

1. Click the "Settings" ⚙️ at the top right
2. Select the "Provider" tab
3. Click "Add Custom Provider"
4. Fill in:
   • Provider name (any, e.g. "My Relay")
   • API Key
   • Base URL
   • Supported model list (auto-pulled from Base URL)
5. Click "Save"

Step 3: Test Connection

1. After saving, click "Test Connection"
2. MBAgent will send a minimal request to verify
3. "Connection successful" displayed → ready to use

IMPORTANT

API Key security: MBAgent's API Key is stored by default in the local Credential Manager (Windows Credential Manager), and is not uploaded to Mbbrowser servers. However, please make sure your computer itself is secure.

---

6. Model Capability Details

Several key parameters when choosing a model:

1. Context Window

How much content the model can "read" at once:

Length	Equivalent	Suitable Scenario
8K	~16 A4 pages	Short chat, simple Q&A
32K	~64 pages	Normal chat, code reading
128K	~256 pages	Long doc analysis, complex codebases
200K	~400 pages	Whole book, deep research
1M+	Extreme	Novel volume, gigantic code

TIP

Just enough is enough: 32K for daily chat; 128K for long PDF analysis; 200K+ for research tasks.

2. Multimodal

Models that "see" images can recognize image content in multimodal scenarios:

• ✅ Diagnose errors from screenshots
• ✅ Generate copy from product images
• ❌ Models without image support are text-only

The "Vision" tag listed by Mbbrowser official indicates multimodal capability.

3. Tool Use / Function Calling

All models listed in MBAgent support tool calling — this is the prerequisite for MBAgent to "do things".

4. Reasoning Capability

Some models (e.g. o1, claude-opus-4 extended thinking) support deep reasoning, taking longer to think but with more accurate results. Suited for:

• Complex math problems
• Multi-step planning
• Hard bug troubleshooting

---

7. Price and Quota Management

View Expense Details

1. Click the "Account" button at the top right
2. Select "Expense Details"
3. The Mbbrowser Console displays for each call: model / Token count / cost / time

Cost-Saving Tips

• ✅ Use small models for daily tasks: gpt-5-mini / deepseek-v3 for 80% of tasks
• ✅ Upgrade for complex tasks: Only swap to claude-opus-4 when really needed
• ✅ Close thinking process: In Settings → Preferences, close "Show thinking process" to save Tokens
• ✅ Context compression: Use /compact to compress history for long sessions
• ❌ Don't open multiple sessions simultaneously: Each session consumes Tokens independently

Package Recommendations

Your Usage	Recommended Package
Occasional, just trying	Free member
< 50 chats per day	Basic member
< 200 chats per day	Personal Custom
Heavy usage, enterprise	Personal Unlimited / Team

---

8. FAQ

Q1: I switched Provider but it's still the same model; why is the cost different?

Different Providers price the same model differently:

• Official vendors (e.g. OpenAI) are generally the most expensive
• Mbbrowser official relay has discounts
• Self-built gateway price is set by you

Q2: The model I want isn't in the list — what to do?

Two ways:

1. Not in Mbbrowser official: contact Mbbrowser customer support to request integration
2. Self-built Provider: use "Add Custom Provider" to integrate any OpenAI-compatible interface

Q3: After switching models, are my historical conversations still there?

✅ Yes. History conversations are saved by "session" dimension, independent of model. You can:

• Start a new session in A model
• Switch model
• Start a new session in B model
• The two sessions don't interfere with each other

Q4: Can the same message be sent to multiple models simultaneously?

✅ Yes:

1. Type the message in the input box
2. Hold Ctrl and click multiple models
3. Models will reply side-by-side
4. Easy to compare effects

---

Next: Task Panel Explained → Handle complex multi-step tasks.