Resolution criteria
Resolves to the first calendar date (US Eastern) when Meta makes Llama 4 “Behemoth” model weights publicly downloadable (any official Behemoth checkpoint: base or instruct) via an official channel. Accepted proof includes, but is not limited to:
A model card under Meta’s Hugging Face org (meta-llama) for Behemoth with downloadable files (“Original”/bf16/fp8 or equivalent). (huggingface.co)
The Meta Llama download portal listing Behemoth with working signed URLs after license acceptance. (github.com)
Meta’s official blog or the meta-llama GitHub “llama-models” table updated to include Behemoth with download instructions. The resolution date is the earliest timestamp the live download becomes available. (github.com)
Exclusions: API-only access, private/partner-only access, leaks/third‑party mirrors, or community quantizations without an official Meta Behemoth checkpoint. If Meta releases quantized Behemoth weights first under the official Meta org/license, that counts. If “Behemoth” is renamed, resolve when Meta’s documentation clearly identifies the Llama 4 teacher/frontier model’s weights as available. (reuters.com)
Background
On April 5, 2025, Meta released Llama 4 Scout and Maverick with open‑weight downloads; Behemoth was previewed as the larger “teacher” model. (reuters.com, huggingface.co)
Meta’s GitHub and Hugging Face pages illustrate how Llama 4 weights are distributed (license click‑through or signed URL), providing clear precedents for how a Behemoth release would appear. (github.com, huggingface.co)
Reporting since May 2025 indicates Behemoth was still training and its public release was delayed to fall 2025 or later; as of Sept 2, 2025, no official Behemoth weights have been posted. (computerworld.com, businessinsider.com)
Considerations
Meta’s “public” releases typically require accepting the Llama license and may use signed URLs; this still counts as public access for this market. Third‑party mirrors do not. (github.com)
If Meta staggers variants (e.g., Instruct before Base), resolve on the first official Behemoth checkpoint posted. (github.com)
Traders can monitor: Meta AI blog, meta‑llama GitHub “llama-models,” and the meta‑llama organization on Hugging Face; these were the primary channels for prior Llama 4 weight releases. (github.com, huggingface.co)