Meta’s frontier model pipeline now has names. Reports attributed to Axios, corroborated by SiliconAngle and TechZine.eu, confirm Meta is developing Avocado, a large language model, and Mango, a multimedia file generator. Both are expected to launch in 2026, according to those reports. Neither is released yet.
The development status of each model differs. Reports indicate the first stage of Avocado’s development is complete. Mango’s status is less clearly defined in available sourcing. The expected 2026 timeline should be treated as a reported target, not a firm commitment, a cross-reference to The New York Times surfaced a headline referencing performance concerns and a possible rollout delay for a Meta AI model, though that delay has not been independently verified for this brief and is flagged as a coverage gap for follow-up.
The open-source question is where the story splits.
Multiple sources, citing Axios as the original reporting, say Meta plans to release open-source versions of both Avocado and Mango. That would be consistent with Meta’s LLaMA strategy, which has made open-weight releases a central part of its AI positioning. Open-source distribution drives developer adoption, builds ecosystem goodwill, and creates competitive pressure on closed-model providers.
But separate reporting tells a different story. Other sources indicate Meta may be reconsidering that strategy and moving toward predominantly closed-source distribution. A third characterization, that Meta will still release some open-source models, just not all of them, adds a middle position to the dispute. The open-source commitment for Avocado and Mango specifically is not settled. It’s a reported plan with active contrary reporting.
That ambiguity matters for the developer community currently building on LLaMA-family models. Those teams are making decisions about infrastructure, fine-tuning pipelines, and deployment architecture based partly on expectations about Meta’s ongoing open-source posture. A shift, even a partial one, would change the calculus. It’s worth watching whether Meta makes a direct statement about Avocado and Mango’s distribution model before or alongside their release.
What’s confirmed: two new frontier models exist and are in active development. What’s contested: the distribution model. What’s unknown: specific capabilities, release dates firmer than “2026,” and whether any delays have affected the timeline.
For practitioners tracking the open-source AI landscape, this brief connects to existing hub coverage of the open-source versus closed-source tension playing out across major AI providers this spring. Avocado and Mango are the next data point in that pattern, one that will carry more weight once Meta’s actual release decisions become clear.