AI Potluck
Model components / Fine-tuned / chat models

Zephyr

Hugging Face

Hugging Face H4's open chat model line. The flagship zephyr-7b-beta is a DPO fine-tune of Mistral 7B trained with the open alignment-handbook recipe on the public UltraChat and UltraFeedback datasets, making the whole run reproducible. The line also includes a 141B mixture-of-experts ORPO variant.

Scored on zephyr-7b-beta. A late-2023 model, so its capability reflects that generation. Verified 2026-08-13 via the model card and the alignment-handbook repository.

Openness

5 high confidence
5.0
weights
open(MIT
data
open(UltraChat_200k + UltraFeedback_binarized, both MIT, ungated)
code
open(alignment-handbook DPO recipe, Apache-2.0)
license
MIT(OSI)

Genuinely fully open: MIT weights, both DPO training datasets public and MIT-licensed, and end-to-end training code (the alignment handbook) under Apache-2.0 that reproduces zephyr-7b-beta. That is open source at the top of the scale. One caveat that does not change the class: the upstream UltraChat and UltraFeedback sets were partly generated with OpenAI model outputs, though the released Hugging Face artifacts are themselves open and MIT.

Adoption

3 high confidence
3.0

The four declared artifacts sum to 232,148 downloads in the trailing 30 days (HuggingFaceH4/zephyr-7b-beta 142,355; HuggingFaceH4/zephyr-orpo-141b-A35b-v0.1 91; HuggingFaceH4/ultrachat_200k 73,327; HuggingFaceH4/ultrafeedback_binarized 16,375), which bands at level 3 (100K-1M) on the software and model adoption scale.

Capability

2 high confidence
2.0

Notable late-2023 capability (a 7B beating much larger chat models on MT-Bench at release), but weak by 2026 standards (GSM8K 12.74, DROP 9.66 on the leaderboard). Low-tier on a scale where 5 = current frontier chat. Scored 2.

Verified 2026-08-13