New Models Added: GPT-5.3-Codex, GPT-Audio-1.5, Qwen3.5 Flash, Qwen3-Coder-Next, and Seedream 5.0

Date: 2026-02-25 / (1404-12-06)

Summary

We announce the availability of six new models across three providers: OpenAI's GPT-5.3-Codex (the most capable agentic coding model) and GPT-Audio-1.5 (the best voice model for audio I/O), Alibaba's Qwen3.5-Flash, Qwen3-Coder-Next, and Qwen3.5-35B-A3B (next-generation efficient models), and BytePlus's Seedream 5.0 (advanced image generation with Chain of Thought reasoning). These models bring state-of-the-art capabilities in agentic coding, audio processing, efficient inference, and image generation.

---

Details

OpenAI

GPT-5.3-Codex

GPT-5.3-Codex (gpt-5.3-codex) is OpenAI's most capable agentic coding model to date. It advances both the frontier coding performance of GPT-5.2-Codex and the reasoning capabilities of GPT-5.2, together in one model, which is also 25% faster. Documentation

Key Features:

• State-of-the-Art Coding: Sets new industry highs on SWE-Bench Pro and Terminal-Bench
• 400K Context Window: Handle large codebases and extensive documentation
• 128K Max Output Tokens: Generate comprehensive code solutions
• Reasoning Token Support: Configurable effort levels (low, medium, high, xhigh)
• Vision Support: Process images for visual code understanding
• Interactive Collaboration: Provides frequent updates during long-running tasks
• Self-Improving: First model instrumental in creating itself
• Endpoint Support: Available on v1/responses

Pricing:

Model	Input	Cached Input	Output
gpt-5.3-codex	$1.75/1M tokens	$0.175/1M tokens	$14.00/1M tokens

GPT-Audio-1.5

GPT-Audio-1.5 (gpt-audio-1.5) is OpenAI's best voice model for audio in, audio out with Chat Completions. It's optimized for real-time interaction with low latency and natural-sounding speech. Documentation

Key Features:

• Multimodal I/O: Text and audio inputs, text and audio outputs
• Low Latency: Optimized for real-time conversational AI
• Natural Speech: Smoother, more conversational audio output
• Function Calling: Support for interactive, tool-driven applications
• 128K Context Window: Handle extensive conversations
• 16K Max Output Tokens: Generate comprehensive responses
• Endpoint Support: Available on v1/chat/completions

Pricing:

Model	Text Input	Cached Input	Audio Input	Text Output	Audio Output
gpt-audio-1.5	$2.50/1M tokens	$1.25/1M tokens	$32.00/1M tokens	$10.00/1M tokens	$64.00/1M tokens

Alibaba (Qwen)

Qwen3.5-Flash

Qwen3.5-Flash (qwen3.5-flash) is the hosted version of Qwen3.5-35B-A3B with production features including 1M context length and official built-in tools. It's designed for extreme inference efficiency using a hybrid architecture. Documentation

Key Features:

• Hybrid Architecture: Gated Delta Networks + Sparse Mixture-of-Experts for high-throughput inference
• Native Multimodal: Process text, images, and videos natively
• 1M Context Window: Extended context for long-form content
• 201 Languages: Expanded language and dialect support
• Built-in Tools: Official adaptive tool use support for agentic workflows
• Cache Creation Support: Efficient caching with tiered pricing

Pricing:

Model	Input	Cache Creation	Cached Input	Output
qwen3.5-flash	$0.10/1M tokens	$0.125/1M tokens	$0.01/1M tokens	$0.40/1M tokens

Qwen3-Coder-Next

Qwen3-Coder-Next (qwen3-coder-next) is an open-weight 80B parameter Mixture-of-Experts model designed specifically for coding agents and local development with only 3B activated parameters. Documentation

Key Features:

• Super Efficient: 80B total / 3B activated parameters - performance comparable to models 10-20x larger
• Advanced Agentic Capabilities: Excels at long-horizon reasoning, complex tool usage, and error recovery
• 256K Context Window: Analyze large code repositories
• IDE Integration: Seamless integration with CLI/IDE platforms (Claude Code, Qwen Code, Cline, Trae, etc.)
• Non-Thinking Mode: Optimized for rapid, real-time code generation

Pricing:

Model	Input	Cached Input	Output
qwen3-coder-next	$0.30/1M tokens	$0.15/1M tokens	$1.50/1M tokens

Qwen3.5-35B-A3B

Qwen3.5-35B-A3B (qwen3.5-35b-a3b) is the open-weight version of Qwen3.5-Flash, featuring 35B total parameters with 3B activated for efficient inference. Documentation

Key Features:

• Hybrid Architecture: Gated Delta Networks + Sparse MoE for efficient inference
• Native Vision-Language: Unified multimodal foundation
• 262K Native Context: Extensible up to 1M tokens
• Strong Coding: 69.2% on SWE-bench Verified
• Agent Capabilities: 81.2% on TAU2-Bench

Pricing:

Model	Input	Cached Input	Output
qwen3.5-35b-a3b	$0.25/1M tokens	$0.12/1M tokens	$2.00/1M tokens

BytePlus (ByteDance)

Seedream 5.0

Seedream 5.0 (seedream-5-0-260128) is ByteDance's latest high-performance image generation model with Chain of Thought reasoning for improved spatial logic and physics compliance. Documentation

Key Features:

• Intelligent Reasoning: Chain of Thought (CoT) for improved spatial logic and physics compliance
• Web-Enabled Context: Connect to live web searches for current events and products
• Multi-Round Editing: Conversational image editing with iterative adjustments
• Enhanced Text Rendering: Improved capabilities for rendering text in images (posters, logos) in multiple languages
• High Resolution: Generate images up to 4K resolution
• Batch Consistency: Stronger subject and layout retention across generations
• Fast Generation: 2-3 seconds per image
• Endpoint Support: Available on v1/images/generations and v1/images/edit

Pricing:

Model	Output
seedream-5-0-260128	$35.00/1M tokens (~$0.035 per image)

---

API Request/Response Examples

GPT-5.3-Codex Example

curl https://api.avalai.ir/v1/responses \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "gpt-5.3-codex",
    "input": "Refactor this codebase to implement dependency injection and add comprehensive test coverage.",
    "reasoning": {"effort": "high"}
  }'

GPT-Audio-1.5 Example

curl https://api.avalai.ir/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "gpt-audio-1.5",
    "messages": [
      {
        "role": "user",
        "content": "Tell me a short story about a robot learning to paint."
      }
    ],
    "modalities": ["text", "audio"],
    "audio": {
      "format": "mp3",
      "voice": "nova"
    }
  }'

Qwen3.5-Flash Example

curl https://api.avalai.ir/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "qwen3.5-flash",
    "messages": [
      {
        "role": "user",
        "content": "Design a comprehensive microservices architecture for an e-commerce platform."
      }
    ],
    "extra_body": {"enable_thinking": false}
  }'

Qwen3-Coder-Next Example

curl https://api.avalai.ir/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "qwen3-coder-next",
    "messages": [
      {
        "role": "user",
        "content": "Write a quick sort algorithm with comprehensive error handling."
      }
    ],
    "max_tokens": 4096,
    "temperature": 1.0,
    "extra_body": {"enable_thinking": false}
  }'

Seedream 5.0 Example

curl https://api.avalai.ir/v1/images/generations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "seedream-5-0-260128",
    "prompt": "Professional product photography of a luxury watch with text 'PREMIUM QUALITY', dramatic lighting, 4K resolution",
    "size": "4K",
    "response_format": "url",
    "sequential_image_generation": "disabled",
    "watermark": false
  }'

---

SDK Usage Examples

Bash

# GPT-5.3-Codex via Responses API
curl https://api.avalai.ir/v1/responses \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "gpt-5.3-codex",
    "input": "Design and implement a REST API for a task management system.",
    "reasoning": {"effort": "medium"}
  }'

Python

from openai import OpenAI

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

# GPT-5.3-Codex for agentic coding
response = client.responses.create(
    model="gpt-5.3-codex",
    input="Design and implement a REST API for a task management system.",
    reasoning={"effort": "medium"},
)

print(response.output_text)

# Qwen3.5-Flash for general tasks
completion = client.chat.completions.create(
    model="qwen3.5-flash",
    messages=[
        {
            "role": "user",
            "content": "Explain quantum computing in simple terms.",
        }
    ],
    extra_body={"enable_thinking": False},
)

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

# Seedream 5.0 for image generation
image_response = client.images.generate(
    model="seedream-5-0-260128",
    prompt="A futuristic cityscape at sunset with flying cars",
    size="4K",
    response_format="url",
    extra_body={"sequential_image_generation": "disabled", "watermark": False},
)

print(f"Generated image: {image_response.data[0].url}")

JavaScript

import { OpenAI } from "openai";

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

// GPT-5.3-Codex for agentic coding
const response = await client.responses.create({
  model: "gpt-5.3-codex",
  input: "Design and implement a REST API for a task management system.",
  reasoning: { effort: "medium" },
});

console.log(response.output_text);

// Qwen3-Coder-Next for coding tasks
const completion = await client.chat.completions.create({
  model: "qwen3-coder-next",
  messages: [
    {
      role: "user",
      content: "Implement a binary search tree with insert, delete, and search operations.",
    },
  ],
  max_tokens: 4096,
  temperature: 1.0,
});

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

// Seedream 5.0 for image generation
const imageResponse = await client.images.generate({
  model: "seedream-5-0-260128",
  prompt: "A futuristic cityscape at sunset with flying cars",
  size: "4K",
  response_format: "url",
});

console.log(`Generated image: ${imageResponse.data[0].url}`);

---

Use Cases

GPT-5.3-Codex

• Agentic Coding: Long-horizon autonomous development tasks
• Large Codebase Refactoring: Handle complex multi-file changes
• Web Development: Build complete applications with games and interactive features
• Debugging: Self-diagnose and fix code issues
• Research Acceleration: Accelerate development workflows

GPT-Audio-1.5

• Voice Agents: Build responsive, human-like voice bots
• Real-time Translation: Efficient live translation
• Interactive Assistants: Conversational interfaces with low latency
• Customer Support: Natural voice-based customer service

Qwen3.5-Flash & Qwen3.5-35B-A3B

• Long-Context Tasks: Process extensive documents and conversations
• Multimodal Agents: Vision and text understanding in unified workflows
• Multilingual Applications: Support for 201 languages and dialects
• Cost-Effective Inference: High-quality outputs with minimal compute

Qwen3-Coder-Next

• Local Development: Efficient coding assistance on consumer hardware
• IDE Integration: Real-time code generation in development environments
• Agentic Workflows: Tool usage and error recovery for complex tasks
• Code Review: Analyze large repositories with 256K context

Seedream 5.0

• E-commerce: Product visualization with text rendering
• Marketing: Professional visuals with enhanced typography
• Architectural Rendering: Accurate material textures and lighting
• UI/UX Mockups: Design prototypes with precise control

---

Related Links

• OpenAI Models Documentation
• Alibaba Models Documentation
• BytePlus Models Documentation
• Reasoning Models Guide
• Image Generation API
• Audio API Reference
• Rate Limits and Pricing