AI model/voice model | Openai goes open source: “Very, very good model” comes without restrictions

Published on: April 1, 2025 / update from: April 1, 2025 - Author: Konrad Wolfenstein

AI model/voice model | Openai goes Open Source: “Very, very good model” comes without restrictions - Image: Xpert.digital

Openai plans open source model without restrictions

Openai announces a new open source language model

Openai plans to publish a new, powerful open source language model with improved skills in the field of logical thinking in the coming months. This announcement marks a significant change of strategy for the company, which has not published a fully open model since the publication of GPT-2. CEO Sam Altman describes the upcoming model as “very, very good model” and emphasizes the intention to provide it without restrictive terms of use.

Suitable for:

Comparative analysis of the leading AI models: Google Gemini 2.0, Deepseek R2 and GPT-4.5 from Openaai

The return to open source roots

After years of concentration on proprietary models such as GPT-3 and GPT-4, this announcement represents a remarkable turnaround in Openai's strategy. The company had largely distanced itself from open source with the publication of GPT-3 and justified this with security concerns. Now CEO Sam Altman admits that the recently closed attitude towards open source could have been problematic.

In a recent Reddit Q&A, Altman said that he personally believes that Openai had to “develop another open source strategy”, even if this is currently not the highest priority of the company. He added that Openai would continue to produce better models, but with a lower lead than in previous years.

This change of course comes at a time when the competitive pressure on Openai is increasing by other AI developers such as Deepseek, Meta, Alibaba and Mistral, which have already successfully launched open models. In particular, the success of the R1 model from Deepseek and Metas Llama family is seen as a possible catalyst for Openai's decision.

Suitable for:

With its AI model R1-OMNI, Alibaba Openaai & Deepseek attacks: R1-OMNI recognizes emotions in videos & describes details

Developer participation and community engagement

In order to make the new model as useful as possible, Openai is actively looking for dialogue with developers, researchers and the wider community. The company has published a feedback form on its website, in which questions such as "What would you want to see in an open white model from Openaai?" and "What open models have you used in the past?" be placed.

In addition, Openaai plans a number of developer events that start in San Francisco and will be continued later in Europe and the Asia-Pacific region. With these events, developers should be able to test early versions of the model and give feedback directly.

Brad Lightcap, COO of Openaai, explains that the decision for an open source model is primarily based on the strong demand of developers as well as business and government customers. "While we will continue to offer top models via our API and Chatgpt, there are many scenarios in which APIs alone do not offer developers full flexibility," said LightCap.

Special features and terms of use

The new model should have special skills in the area of logical conclusion (Reasoning), similar to the O3 mini model from Openaai. It is also noteworthy that Altman emphasizes that the model will be published without restrictive terms of use.

In an obvious swipe on Meta, he said: "We will not introduce a silly restriction, such as that the service may not have more than 700 million monthly active users." This limit corresponds exactly to the restriction of usage of METAS LLAMA model. While Llama is criticized not to be an open source, since Meta does not open the training data in addition to the license restriction, it remains to be seen whether Openaai will only publish the model weights or actually provide all components for the complete replica of the system including the training data.

Security aspects and evaluation

Before the publication, the model is evaluated according to the Openai Preparedness framework. The specific risks of an open source model are particularly taken into account, such as possible enemy fine-tunes by third parties.

Johannes Heidecke von Openaai emphasizes that the company carries out security tests, both internally and with external experts. Openaai will not publish any models from which catastrophic risks could arise, said Heidecke. This additional care is necessary because it is known that the model could be modified after the publication of third parties.

Effects on the AI landscape

Openai's decision to publish an open source model could have far-reaching effects on the AI landscape. In an industry that is increasingly shaped by competition, the winner of the open source race could have the advantage of making proprietary offers superfluous and becoming the dominant platform among developers, similar to how Google has succeeded with Android.

For developers, a high-quality open source model from Openai means more flexibility in special use cases such as specialized fine-tunes, customizable latency times, local installations or applications that require full control over the data. These scenarios are currently not optimally served by APIS alone.

Openai relies on open source: a turning point in the AI strategy?

Openai's announcement to publish a new open source language model with Reasoning capabilities marks significant change in the company's strategy. In view of the increasing competitive pressure and the growing demand for open models, Openai seems to re-evaluate its position in the AI market.

The coming months will show whether the new model can do justice to the high expectations and how it does it compared to other open source alternatives. The active involvement of the developer community and the waiver of restrictive terms of use could take an important step to put its original mission - democratization of artificial intelligence - more in the foreground.

Suitable for: