Meta has a roadmap of new Llama releases bringing multimodality, more multilinguality and bigger context windows.

The test is whether Llama releases after 18 April 2024 delivered three features: multimodality (image or other non-text input), more multilinguality (official support for more languages than Llama 3's English focus), and bigger context windows (longer than Llama 3's 8K tokens). Delivery counts when publicly released models, such as Llama 3.1, 3.2 or 4, have these features as documented in Meta's model cards or release announcements. True: all three were delivered in subsequent releases. Mixed: two of the three were delivered. Mostly false: only one was delivered. False: none was delivered, or the roadmap was abandoned without any of them.

Set before the research began.

1. E1 Llama 3.1 - 405B, 70B & 8B with multilinguality and long context Hugging Face · 23 July 2024

   “All the variants support a context length of 128K tokens and 8 languages, including English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai.”

2. E2 Announcing Llama 3.1 405B, 70B, and 8B models from Meta in Amazon Bedrock | Amazon Web Services Amazon Web Services · 23 July 2024

   “All Llama 3.1 models support a 128K context length (an increase of 120K tokens from Llama 3)”

3. E3 llama-models/models/llama3_2/MODEL_CARD.md at main · meta-llama/llama-models GitHub (Meta Llama repository) Own record

   “Supported Languages: English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai are officially supported. Llama 3.2 has been trained on a broader collection of languages than these 8 supported languages.”

4. E4 Unmatched Performance and Efficiency | Llama 4 Meta (dev.meta.ai) Own record

   “All Llama 4 models are designed with native multimodality, leveraging early fusion that allows us to pre-train the model with large amounts of unlabeled text and vision tokens”

Independent launch coverage of Llama 3.1, released after the claim date, says all variants support a 128K-token context length (up from Llama 3's 8K) and eight officially supported languages (E1, E2). Meta's Llama 3.2 model card confirms the eight officially supported languages (E3). Meta's Llama 4 release page says all Llama 4 models are natively multimodal and were pre-trained on text and vision tokens (E4). Under the criteria, a release announcement or model card from Meta counts as documentation, so all three promised features were delivered. There was no deadline, so on_time is null.