For the first time in about five years, Openai is preparing to release an AI system that is truly “open.” This means that it cannot be downloaded because it is not ranked behind the API. TechCrunch reported Wednesday that Openai is aiming for an early summer launch, targeting better performance than open models from Meta and Deepseek.
Beyond Benchmark Performance, Openai may have important features in its sleeves. This can make an open “inference” model highly competitive, TechCrunch learned.
Company leaders are discussing plans to enable open models to connect to Openai’s cloud-hosted model, and are able to answer complex queries well. In a recent meeting with developers in the open source AI community, Openai CEO Sam Altman described this ability as a “handoff,” according to one of the sources.
Openai did not respond to TechCrunch’s request for comment.
As the feature is described, if the feature becomes an open model, you will be able to call the Openai API and access other larger models in the company for substantial calculation lifts. It is unclear whether open models will have access to some of the many tools that OpenAI models can use, such as web search and image generation.
According to sources, the idea for the Handoff feature was proposed by the developer in one of Openai’s recent developer forums. The proposal appears to have gained traction within the company. Openai will host a series of community feedback events with developers to help shape upcoming open model releases.
Local models that can take advantage of more powerful cloud systems are reminiscent of Apple Intelligence, Apple’s suite of AI capabilities that use a combination of device models and models running in “private” data centers. Openai can benefit in obvious ways. In addition to generating incremental revenue, handoffs allow more members of the open source community to rope into the company’s premium ecosystem.
There is a lot of unknowns about how Openai’s open model handoff functionality works, including what the pricing and fee limits are. The model is in its early stages and its aspects may not change or realize.
Rather than reusing old models for open releases, Openai is in the process of training new models from scratch, sources tell TechCrunch. One source predicts that Open Model is performing Openai’s O3, but it is expected to outperform Deepseek’s R1 inference model in certain benchmarks.