Close Menu
  • AI
  • Content Creation
  • Tech
  • Robotics
AI-trends.todayAI-trends.today
  • AI
  • Content Creation
  • Tech
  • Robotics
Trending
  • A Group of Researchers from Columbia and CertiK Introduce TensorLean for Formal Verification of GPU Kernels
  • Collectively AI Discloses Immediate Leak After Coding Agent Uncovered Credential – Unite.AI
  • Dragos Finds Immediate Leaks and Stolen Keys in AI Provide Chains
  • Lean and Mathlib Communities Certify Claude-Assisted Riemann Proof – Unite.AI
  • AWS Routes HyperPod Inference Throughout Areas From One Endpoint – Unite.AI
  • DailyAI Briefing: A Lean-checked Riemann proof, plus 4 extra
  • Anthropic Shares Drop 8.5% in Second Session Throughout AI Selloff – Unite.AI
  • OpenAI Ships GPT-Stay-1.1 for Voice Brokers, Cuts Idle Value to $0.02/Min
AI-trends.todayAI-trends.today
Home»AI»OpenAI has released its first Open-Weight models since GPT-2

OpenAI has released its first Open-Weight models since GPT-2

AI By Gavin Wallace05/08/20253 Mins Read
Facebook Twitter LinkedIn Email
Anthropic’s New Model Excels at Reasoning and Planning—and Has the
Anthropic’s New Model Excels at Reasoning and Planning—and Has the
Share
Facebook Twitter LinkedIn Email

OpenAI has just released its first models with open weights in more than five years. These two models can be run on local devices by consumers and fine-tuned to specific needs. The language models gpt-oss-20b and gpt-oss 120b can be run locally on consumer devices, fine tuned for specific purposes. OpenAIThe company is now moving away from focusing solely on its proprietary AI releases and towards more accessible, open-source models.

“We’re excited to make this model, the result of billions of dollars of research, available to the world to get AI into the hands of the most people possible,” OpenAI CEO Sam Altman said in an email statement. Hugging Face hosts both gpt’s oss-120b as well as gpt’s oss-20b for free. AI tools. OpenAI released its last model with an open weight in 2009. GPT-2Back in 2019

An open-weighted model differs from other models in that it is lighter. “weights” The internal parameters are available to anyone, so they can get a better idea about how the system works. Greg Brockman, the cofounder of OpenAI, sees it as a way to complement their proprietary models. “complementary” The application programming interface, which is currently being used by many developers. “Open-weight models have a very different set of strengths,” Brockman said in a press briefing. And unlike ChatGPT you can also run the gpt oss without any internet access and even behind a fire wall.

These models are both gpt oss chain-of-thought reasoning What are the best approaches? OpenAI first deployed in its o1 model last fall. Instead of just providing an answer, this method has the generative AI tool go through multiple stages to address a question. Text-only models can’t be multimodal but are able to browse the web and call models in the cloud for help. They also execute code as well as navigate software. AI agent. The smaller model gptoss-20b can be installed locally in a device that has more than 16 GB memory.

OpenAI has released two new models under its OpenAI brand. Apache 2.0 licenseA popular option for open-weighted models. Apache 2.0 allows models to be redistributed and used in other software. Alibaba releases open-weight models Qwen The word “as” is also used to describe the term. Mistral Apache 2.0 is also supported.

The release of the open-weight models, which were first announced publicly in March for additional safety testing, was originally delayed. The release of an open weight model could be more hazardous than the closed version, as it would remove any barriers to who is allowed to use the tool. Anyone can also try and fine-tune their version of GPT-OSS for non-intended uses.

The startup, in addition to the usual evaluations OpenAI runs on their proprietary models to determine how they could be abused by others, customized the option of open-weight to test its potential misuse. “bad actor” Downloading the software is up to you. “We actually fine-tuned the model internally on some of these risk areas,” Eric Wallace is a researcher in safety at OpenAI. “and measured how high we could push them.” OpenAI’s test results showed that OpenAI’s open-weight model didn’t reach high levels of risk as determined by its preparedness framework.

AI algorithms artificial intelligence meta open source openai
Share. Facebook Twitter LinkedIn Email
Avatar
Gavin Wallace

Related Posts

Dragos Finds Immediate Leaks and Stolen Keys in AI Provide Chains

22/11/2026

DailyAI Briefing: A Lean-checked Riemann proof, plus 4 extra

21/11/2026

Common Tech Bros Are Fearful About Honeypots

21/11/2026

Cops Are Pushing Extraordinarily Corny ‘Grand Theft Auto 6’ Slop

21/11/2026
Top News

Couples retreat with 3 chatbots that are AI and humans who love them

A New AI Tool from Flock is Powerful for Police. We Found Its Code

Anthropic’s IPO Prospectus Reveals What’s Below the Hood

My father wants to age in place. AI Will Be Watching

Disney Simply Threw a Punch in a Main AI Combat

Load More
AI-Trends.Today

Your daily source of AI news and trends. Stay up to date with everything AI and automation!

X (Twitter) Instagram
Top Insights

Alibaba Qwen Introduces QwenGyre: Elastic Reinforcement Studying RL for XLong-Horizon Brokers

12/10/2026

What I learned from using food-tracking apps was more than I expected

20/03/2026
Latest News

A Group of Researchers from Columbia and CertiK Introduce TensorLean for Formal Verification of GPU Kernels

22/11/2026

Collectively AI Discloses Immediate Leak After Coding Agent Uncovered Credential – Unite.AI

22/11/2026
X (Twitter) Instagram
  • Privacy Policy
  • Contact Us
  • Terms and Conditions
© 2026 AI-Trends.Today

Type above and press Enter to search. Press Esc to cancel.