OpenAI’s next open AI model could arrive with a feature that blurs the line between local software and cloud-based AI services. The company is preparing a model that would be available for download at no cost and not gated behind an API, while also exploring a way for that model to call OpenAI’s larger cloud-hosted systems when a task needs more computing power.
A new open model with a cloud option
TechCrunch reported that OpenAI is aiming for an early summer launch of its first truly “open” AI system in roughly five years. The model is expected to be downloadable rather than available only through an API, a distinction that matters to developers who want to run AI systems locally or build with fewer platform constraints.
The company is also targeting performance that would exceed open models from Meta and DeepSeek. The model is described as an open “reasoning” model, meaning its value will likely be judged not only by general output quality but also by how well it handles complex tasks that require step-by-step problem solving.
The most notable detail is not just that OpenAI is preparing an open release. According to two sources familiar with the matter, company leaders have discussed enabling the open model to connect to OpenAI’s cloud-hosted models when a query is too demanding for the local model alone.
During a recent meeting with developers in the open source AI community, OpenAI CEO Sam Altman described the capability as a “handoff,” according to one source cited by TechCrunch. OpenAI did not respond to TechCrunch’s request for comment.
How the “handoff” could work
If the feature is included as described, the open model would be able to make calls to the OpenAI API and use the company’s other, larger models for a substantial computational lift. In plain terms, the downloaded model could handle some work locally, then pass harder requests to cloud-hosted OpenAI models when needed.
That would make the model different from a purely local system. Developers could still get the benefits of a model that is not initially gated behind an API, but they might also have a path to stronger performance on difficult prompts through OpenAI’s existing cloud infrastructure.
Several important details remain unresolved. TechCrunch reported that it is unclear whether the open model would be able to access tools used by OpenAI’s models, such as web search and image generation. It is also unknown what pricing and rate limits would apply if the handoff feature launches.
Those unknowns matter because they shape how developers would actually use the system. A handoff feature could be powerful, but the practical value would depend on when the model chooses to call the cloud, how much those calls cost, and how predictable the limits are for real applications.
Why developers may care
The idea reportedly came from a developer during one of OpenAI’s recent developer forums. According to a source cited by TechCrunch, the suggestion appears to have gained traction inside the company. OpenAI has been holding community feedback events with developers as it shapes the open model release.
That origin is important. The handoff concept reflects a common tension in AI development: local models can be easier to download, inspect, and integrate, while larger cloud systems can offer more computational strength for difficult requests. A model that can move between those two modes could appeal to developers who want both local access and optional cloud assistance.
TechCrunch compared the concept to Apple Intelligence, Apple’s suite of AI capabilities that combines on-device models with models running in “private” data centers. The comparison is not exact in every detail, but it points to the same broad design pattern: keep some AI work close to the user or device, and send more demanding tasks to larger systems when necessary.
For OpenAI, the benefits could be direct. TechCrunch noted that a handoff could generate incremental revenue and bring more members of the open source community into the company’s premium ecosystem. That would give OpenAI a way to release a no-cost downloadable model while still creating a path back to paid cloud services.
Performance expectations and open questions
OpenAI is training a new model from scratch rather than repurposing an older one for the open release, sources told TechCrunch. One source said the open model is expected to underperform OpenAI’s o3, but outperform DeepSeek’s R1 reasoning model on certain benchmarks.
That positioning suggests OpenAI is not presenting the open model as its strongest overall system. Instead, the model may be designed to compete with leading open reasoning models while leaving room for OpenAI’s larger cloud-hosted models to handle more demanding workloads.
The cloud handoff feature is still uncertain. TechCrunch reported that the model is in the early stages, and parts of the plan could change or not happen at all. That means developers should treat the feature as a reported direction rather than a confirmed product specification.
The key open questions include:
- Whether the handoff feature will be included in the final release.
- Which OpenAI cloud-hosted models the open model could access.
- Whether tools such as web search and image generation would be available.
- How pricing and rate limits would work.
- How the model would decide when a query needs cloud assistance.
Even with those uncertainties, the reported plan shows how OpenAI may try to compete in open AI without fully separating that effort from its cloud business. A downloadable open AI model could attract developers who want local access, while a handoff to the OpenAI API could keep advanced reasoning tasks connected to the company’s larger systems.
If the feature ships, the result would not be a simple local-versus-cloud choice. It would be a model that begins on the developer’s machine but can reach upward when the task demands more power. For OpenAI, that could make its open model more competitive. For developers, the real test will be whether the added capability is useful, affordable, and clear enough to build around.