OpenAI is moving toward a new kind of release: an open language model designed to compete directly with other open reasoning models. The project is still in its early stages, but the direction is becoming clearer through the company’s sessions with the AI developer community.
The upcoming model matters because it would mark OpenAI’s first open language model since GPT‑2. It also arrives at a moment when open models are becoming a serious competitive force, giving developers more room to test, adapt, and in some cases commercialize AI systems outside the usual closed-product path.
A return to open models
Toward the end of March, OpenAI said it intended to release an open language model sometime this year. The plan now appears to be more specific: the company is aiming for an early summer release, according to sources familiar with the proceedings.
Aidan Clark, OpenAI’s VP of research, is leading development of the model. The work is described as being in the very early stages, which means the final shape of the release could still change before it reaches developers.
The model is expected to be a reasoning model along the lines of OpenAI’s o-series models. That detail is important because reasoning models are built around more deliberate problem-solving behavior, rather than simple text generation alone.
OpenAI also wants the model to top benchmarks against other open reasoning models. Benchmark performance is not the whole story for developers, but it is a visible way for AI companies to signal capability and compete for attention in a crowded field.
Why licensing is central
OpenAI is exploring a highly permissive license with few usage or commercial restrictions, according to the same sources. That would directly address one of the recurring tensions around open AI releases: whether a model is open in a practical sense, or only available under conditions that limit what developers can do with it.
Open models such as Llama and Google’s Gemma have been criticized by some in the community for imposing onerous requirements. OpenAI appears to be looking for a path that avoids those complaints.
A permissive license could make the model more attractive to developers who want to experiment, build products, or adapt the system for their own use cases. The source does not specify the final license terms, so the key point is that OpenAI is exploring fewer restrictions, not that a final licensing decision has been announced.
The competitive pressure is clear. Rivals, including Chinese AI lab DeepSeek, have adopted an open approach to launching models. Those competitors make models available to the AI community for experimentation and, in some cases, commercialization.
The competitive backdrop
OpenAI’s move comes after open-model strategies have produced major momentum for other organizations. Meta has invested heavily in its Llama family of open AI models and said in early March that Llama has racked up over 1 billion downloads.
DeepSeek has also built a large worldwide user base and attracted the attention of investors. Those outcomes show why open releases can be strategically powerful: they can spread quickly through the developer community, encourage experimentation, and turn technical access into wider influence.
For OpenAI, the issue is not only whether it can release a capable model. It is whether the company can define an open strategy that fits its position in the market while answering developer demand for more flexible access.
Sam Altman has acknowledged that tension before. During a Reddit Q&A in January, he said, "[I personally think we need to] figure out a different open source strategy," adding, "Not everyone at OpenAI shares this view, and it’s also not our current highest priority […] We will produce better models [going forward], but we will maintain less of a lead than we did in previous years."
That comment frames the upcoming model as more than a single release. It reflects a broader question about how OpenAI balances closed systems, open access, developer trust, and competitive pressure.
What developers may get
Sources said OpenAI intends for the model to be "text in, text out." In plain terms, the release is expected to focus on language input and language output, rather than being presented as a broader multimodal system.
The company also intends for the model to run on high-end consumer hardware. That could make it more accessible to developers who want to work locally, though the source does not define exact hardware requirements.
Another feature under consideration is a toggle that would let developers turn the model’s "reasoning" on or off. That would be similar to reasoning models recently released by Anthropic and others. Because the source describes this as possible, it should be treated as a feature under exploration rather than a confirmed launch detail.
If the release is well-received, OpenAI may follow up with additional models. Those could potentially include smaller models as well. Again, the source presents this as a possible next step, not a fixed roadmap.
Safety and trust will shape the launch
OpenAI’s open model is also being framed around safety work. Altman has said the model will be thoroughly red-teamed and evaluated for safety before release.
The company intends to release a model card for the model, according to sources. That model card would be a thorough technical report showing the results of OpenAI’s internal and external benchmarking and safety testing.
Altman also said in a post on X last month, "[B]efore release, we will evaluate this model according [to] our preparedness framework, like we would for any other model," and added, "[A]nd we will do extra work given that we know this model will be modified post-release."
That last point is especially relevant for an open release. Once a model is available to the broader community, developers may modify it after release. OpenAI appears to be treating that as a safety consideration from the start.
The company enters this launch with scrutiny already attached. OpenAI has raised concerns among some AI ethicists for reportedly rushing safety testing of recent models and for failing to release model cards for others. Altman also stands accused of misleading OpenAI executives about model safety reviews before his brief ouster in November 2023.
For developers, the final judgment will likely rest on several connected questions: how capable the model is, how open the license truly feels in practice, whether it runs well on high-end consumer hardware, and how much detail OpenAI provides about testing. The company is trying to meet the open-model moment, but the details of the release will decide how far that shift really goes.