New Model Added: GLM-4.6 from Z.AI and DeepSeek-V3.2 Upgrade

Date: 2025-10-04

Summary

We announce the addition of GLM-4.6, a cutting-edge AI model from Z.AI optimized for coding and long-context tasks, alongside an automatic upgrade of DeepSeek models to the new DeepSeek-V3.2-Exp base with 50%+ price reduction. GLM-4.6 delivers superior coding performance with a 200K token context window, while DeepSeek-V3.2 introduces efficient sparse attention for enhanced long-context processing.

---

Details

Z.AI

GLM-4.6

We introduce GLM-4.6, the latest model from Z.AI achieving comprehensive enhancements across multiple domains including real-world coding, long-context processing, reasoning, searching, writing, and agentic applications. Documentation

Key Features:

• Context Window: 200K tokens (expanded from 128K) for handling complex agentic tasks
• Advanced Capabilities: Superior coding performance, tool use during inference, function calling, structured outputs
• Maximum Output: 128K tokens
• Reasoning Support: Built-in thinking mode with thinking parameter
• Real-World Performance: Outperforms Claude Sonnet 4 in practical coding tests within Claude Code environment
• Endpoint Support: Available on v1/chat/completions
• Web Search: Integrated web search capabilities at $0.01 per tool call

Pricing Details:

Model	Input	Cached Input	Output	Web Search
glm-4.6	$0.60/1M tokens	$0.11/1M tokens	$2.20/1M tokens	$0.01/call

Performance Highlights:

GLM-4.6 achieves performance on par with Claude Sonnet 4/Claude Sonnet 4.6 across 8 authoritative benchmarks (AIME 25, GPQA, LCB v6, HLE, SWE-Bench Verified), solidifying its position as a top-tier model. In real-world coding tests, GLM-4.6 demonstrates over 30% better token efficiency compared to its predecessor.

Example Usage:

Bash

curl https://api.avalai.ir/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "glm-4.6",
    "messages": [
      {
        "role": "user",
        "content": "Create a responsive navigation bar with dropdown menus using React and Tailwind CSS."
      }
    ],
    "thinking": {
      "type": "enabled"
    },
    "max_tokens": 4096,
    "temperature": 0.6
  }'

Python

from openai import OpenAI

client = OpenAI(api_key="your-avalai-api-key", base_url="https://api.avalai.ir/v1")

response = client.chat.completions.create(
    model="glm-4.6",
    messages=[
        {
            "role": "user",
            "content": "Create a responsive navigation bar with dropdown menus using React and Tailwind CSS.",
        }
    ],
    extra_body={"thinking": {"type": "enabled"}},
    max_tokens=4096,
    temperature=0.6,
)

print(response.choices[0].message.content)

JavaScript

import { OpenAI } from "openai";

const client = new OpenAI({
    apiKey: process.env.AVALAI_API_KEY,
    baseURL: "https://api.avalai.ir/v1"
});

const response = await client.chat.completions.create({
    model: "glm-4.6",
    messages: [
        {
            role: "user",
            content: "Create a responsive navigation bar with dropdown menus using React and Tailwind CSS."
        }
    ],
    thinking: {
        type: "enabled"
    },
    max_tokens: 4096,
    temperature: 0.6
});

console.log(response.choices[0].message.content);

Use Cases:

• AI Coding: Superior performance in Claude Code, Cline, OpenCode, and Roo Code with enhanced frontend aesthetics
• Smart Office: Advanced PowerPoint creation and office automation with clear logical structures
• Translation: Optimized quality for French, Russian, Japanese, Korean, and informal contexts
• Content Creation: Natural expression in novels, scripts, and copywriting with contextual expansion
• Virtual Characters: Consistent tone across multi-turn conversations for social AI applications
• Intelligent Search: Enhanced intent understanding and deep research capabilities

DeepSeek

Automatic Upgrade to DeepSeek-V3.2-Exp

DeepSeek models have been automatically upgraded to the new DeepSeek-V3.2-Exp base model, featuring DeepSeek Sparse Attention (DSA) for faster, more efficient training and inference on long context. This upgrade is seamless and requires no changes to your existing implementation.

What's New:

• DSA Technology: Fine-grained sparse attention with minimal impact on output quality
• Enhanced Efficiency: Improved long-context performance with reduced compute cost
• Performance: On par with V3.1-Terminus across benchmarks
• Price Reduction: Over 50% cost decrease on API pricing

⚠️ Important: Thinking Mode API Changes (Updated December 2025)

DeepSeek has updated their API for thinking mode (deepseek-reasoner). When using tool calls with thinking mode, you must now pass the reasoning_content field back to the API in subsequent requests.

Key Changes:

1. Response Fields: The API now returns both reasoning_content (CoT reasoning) and content (final answer)
2. Multi-turn Conversations: Only pass content from previous turns, not reasoning_content
3. Tool Calls with Thinking Mode: You MUST include reasoning_content in assistant messages when processing tool calls within the same turn

Error if Not Implemented:

Missing reasoning_content field in the assistant message

Quick Fix Example:

# When receiving tool_calls from deepseek-reasoner:
assistant_message = {
    "role": "assistant",
    "content": message.content or "",
    "tool_calls": [...],
    # CRITICAL: Include reasoning_content
    "reasoning_content": message.reasoning_content,
}
messages.append(assistant_message)

For complete documentation and examples in multiple languages, see the DeepSeek Models Documentation.

Official Reference: DeepSeek Thinking Mode - Tool Calls

Updated Pricing:

Model	Previous Input	Previous Output	New Input	New Output	Savings
deepseek-chat	$0.56/1M	$1.68/1M	$0.28/1M	$0.42/1M	50%+
deepseek-reasoner	$0.56/1M	$1.68/1M	$0.28/1M	$0.42/1M	50%+

Note: Cached input pricing also reduced from $0.07/1M to $0.028/1M tokens.

Model Routing:

Your existing API calls to deepseek-chat (for standard chat) and deepseek-reasoner (for reasoning mode) are automatically routed to the new DeepSeek-V3.2-Exp base model. No action required on your part.

Example Usage:

Bash

# Standard chat mode
curl https://api.avalai.ir/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "deepseek-chat",
    "messages": [
      {
        "role": "user",
        "content": "Explain quantum computing in simple terms."
      }
    ]
  }'

# Reasoning mode
curl https://api.avalai.ir/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "deepseek-reasoner",
    "messages": [
      {
        "role": "user",
        "content": "Design a scalable microservices architecture for an e-commerce platform."
      }
    ]
  }'

Python

from openai import OpenAI

client = OpenAI(api_key="your-avalai-api-key", base_url="https://api.avalai.ir/v1")

# Standard chat mode
response = client.chat.completions.create(
    model="deepseek-chat",
    messages=[
        {"role": "user", "content": "Explain quantum computing in simple terms."}
    ],
)

print(response.choices[0].message.content)

# Reasoning mode
reasoning_response = client.chat.completions.create(
    model="deepseek-reasoner",
    messages=[
        {
            "role": "user",
            "content": "Design a scalable microservices architecture for an e-commerce platform.",
        }
    ],
)

print(reasoning_response.choices[0].message.content)

JavaScript

import { OpenAI } from "openai";

const client = new OpenAI({
    apiKey: process.env.AVALAI_API_KEY,
    baseURL: "https://api.avalai.ir/v1"
});

// Standard chat mode
const response = await client.chat.completions.create({
    model: "deepseek-chat",
    messages: [
        {
            role: "user",
            content: "Explain quantum computing in simple terms."
        }
    ]
});

console.log(response.choices[0].message.content);

// Reasoning mode
const reasoningResponse = await client.chat.completions.create({
    model: "deepseek-reasoner",
    messages: [
        {
            role: "user",
            content: "Design a scalable microservices architecture for an e-commerce platform."
        }
    ]
});

console.log(reasoningResponse.choices[0].message.content);

---

Related Links

• Z.AI Models Documentation
• Function Calling Guide
• Reasoning Models Guide
• Rate Limits and Pricing
• Best Practices for Production