According to the latest reports, Meta Platforms is planning to delay the release of its flagship AI model, Behemoth. Originally scheduled for April this year, Behemoth was first postponed to June and is now expected to be delayed until the fall of 2025 or even later. This change has sparked concerns within the company regarding the direction of Meta's multi-billion-dollar investment in the AI field.
Meta's engineers have encountered difficulties in enhancing the capabilities of the Behemoth large language model. Some employees question whether the performance improvement over the previous model is sufficient to support a public release. Although Meta had planned to release Behemoth at its AI conference in April and had already released two smaller models from the Llama AI series prior to the event, the release plan now faces uncertainty once again.
Neither Zuckerberg nor other executives have made specific commitments regarding Behemoth's release date. However, Meta may consider launching a feature-limited version earlier. Nevertheless, Meta's engineers and researchers are concerned about the model's performance, believing it may not meet previous claims.
Meta had confidently claimed that Behemoth outperformed OpenAI, Google, and Anthropic in certain tests. However, insiders stated that the model's performance is constrained by challenges during the training process. A Meta spokesperson declined further comment on this matter.
Meta also plans to spend up to $72 billion on capital expenditures this year, with most funds allocated to achieve Zuckerberg's ambitious goals in the AI field. However, as the release of Behemoth has been postponed, the market has reacted negatively to Meta's stock price, with shares falling more than 3% during Thursday trading. This change also dragged down other tech stocks, including Nvidia and the Nasdaq Composite Index.
Meanwhile, Meta expressed dissatisfaction with the performance of its AI product team, particularly the team developing the Llama models. After Meta's first-generation Llama model was launched in early 2023, most of the researchers involved left, and subsequent model development encountered problems. Although the two Llama models released in April initially performed well, they were later found to differ from the publicly released models.
Meta's senior leadership is disappointed with the performance of the team developing the Llama4 model and attributes the slow progress of Behemoth to this team, thus considering adjustments to its AI product group management. These events indicate that Meta is facing increasing challenges and pressure in the global AI competition.
Key points:
💡 Meta is delaying the release of its flagship AI model, Behemoth, possibly until the fall of 2025.
📉 This change has triggered internal concerns about the direction of its multi-billion-dollar AI investments, with the stock price dropping more than 3%.
🤖 Meta is dissatisfied with the performance of its AI product teams and may make management adjustments.