one year on
Meta releases Llama 3.3 70B Instruct
The open-weight instruction-tuned model, fine-tuned from Llama 3.1 70B, scores 94.8% on GSM8K and 65.9% on MMLU-Pro.
Meta today released Llama 3.3 70B Instruct, an open-weight instruction-tuned model. The model, fine-tuned from Llama 3.1 70B, scores 94.8% on GSM8K and 65.9% on MMLU-Pro. The model logged 745,246 downloads on Hugging Face.
One year later — open only if you can handle spoilers
Llama 3.3 70B became a workhorse for the open-source community throughout early 2025, widely adopted for cost-sensitive deployments. Meta's troubled Llama 4 release the following spring, plagued by benchmark cherry-picking and rapid-fire retractions, only made Llama 3.3 look better in retrospect.
The Weekly Replay · free by email
This week, one year ago — every Sunday.
One email each Sunday: the week's replayed AI news, with the one-year-later annotations included. Written like it's breaking — dated like it isn't.
Free · double opt-in · unsubscribe anytime · privacy