New Models Added: OpenAI's First Open-Source Models and API Performance Upgrades
Date: 2025-08-12
Summary
We are excited to announce the addition of OpenAI's first-ever open-source models, gpt-oss-120b and gpt-oss-20b, now available through Azure AI and AWS Bedrock. This update also includes significant performance enhancements to the AvalAI API, delivering lower latency and improved reliability for a production-grade experience.
Details
Core API Updates
We have rolled out significant updates to our core infrastructure, resulting in a faster and more reliable AvalAI API. Users will benefit from:
- Lower Latency: Reduced response times for faster interactions.
- Improved Performance: Enhanced stability and throughput for production workloads.
- Greater Reliability: A more robust and dependable service.
These improvements are part of our ongoing commitment to providing a best-in-class API experience.
New Open-Source Models from OpenAI
For the first time, OpenAI has released open-source models, and we are making them available to you through our provider partners.
Azure AI
- gpt-oss-120b: OpenAI's most powerful open-weight model, designed for high-end reasoning tasks and fitting within a single H100 GPU. Documentation
AWS Bedrock
- openai.gpt-oss-120b-1:0: The same powerful 120B parameter model, available through AWS Bedrock, offering a competitive pricing structure. Documentation
- openai.gpt-oss-20b-1:0: A medium-sized 20B parameter model, perfect for use cases requiring low latency and high efficiency. Documentation
Key Features of the New Models
Both gpt-oss-120b and gpt-oss-20b come with a range of powerful features:
- Permissive Apache 2.0 License: Freedom to build, customize, and deploy commercially without restrictive licensing.
- Configurable Reasoning Effort: Adjust the model's reasoning effort (low, medium, high) to match your needs.
- Full Chain-of-Thought: Complete transparency into the model's reasoning process for better debugging and trust.
- Fine-Tunable: Adapt the models to your specific tasks with parameter-efficient fine-tuning.
- Agentic Capabilities: Native support for function calling, web browsing, code execution, and structured outputs.
Usage Example
from openai import OpenAI
client = OpenAI(api_key="your-avalai-api-key", base_url="https://api.avalai.ir/v1")
completion = client.chat.completions.create(
model="gpt-oss-120b", # or "openai.gpt-oss-120b-1:0"
messages=[
{
"role": "user",
"content": "Explain the significance of the Apache 2.0 license for open-source models.",
}
],
)
print(completion.choices[0].message.content)import { OpenAI } from "openai";
const client = new OpenAI({
apiKey: process.env.AVALAI_API_KEY,
baseURL: "https://api.avalai.ir/v1",
});
const completion = await client.chat.completions.create({
model: "gpt-oss-120b", // or "openai.gpt-oss-120b-1:0"
messages: [
{
role: "user",
content: "Explain the significance of the Apache 2.0 license for open-source models.",
},
],
});
console.log(completion.choices[0].message.content);Pricing and Availability
Pricing for these models is determined by the provider. The gpt-oss-120b model on Azure AI is priced differently than the openai.gpt-oss-120b-1:0 version on AWS Bedrock, with the latter offering a more cost-effective option. Please refer to the individual model documentation for detailed pricing information.