NousResearch Releases Arete-27B, Cuts Flagged Propaganda From 22% to 1%
NousResearch released Arete-27B, a Qwen3.8-27B base model post-trained to reinforce Western values including freedom of expression, democracy and privacy. According to the organization, fine-tuning on roughly 53 million tokens cut its propaganda rate from 22% to 1% on an 803-prompt evaluation across seven value categories, with no significant Hermes Bench tradeoff. NousResearch says the weights, training data and evaluation are publicly available. The release is framed against Chinese open models, which the organization describes as currently the strongest but carrying state-aligned behavior.
IllustrationEditorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Reporting timeline
NousResearch Releases Arete-27B, a Qwen3.8-27B Fine-Tune to Reinforce Western Values
AI research group NousResearch has released Arete-27B, a fine-tuned version of the Qwen3.8-27B model, according to a post shared by DataChaz on X. The poster said the model was tuned to reinforce Western values, and praised the work, writing that they "really like what @NousResearch has done here." The post does not detail the training data, evaluation results, licensing terms, or benchmark performance of the release. No independent confirmation of the model's capabilities or availability was included in the source item. The announcement frames the release around value alignment rather than technical performance claims.
NousResearch Releases Arete-27B, Qwen3.8-27B Fine-Tuned to Reinforce Western Values
NousResearch has released Arete-27B, a fine-tune of Qwen3.8-27B designed to reinforce Western values including freedom of expression, democracy and privacy. According to the post, on the team's 803-prompt evaluation, answers flagged as propaganda dropped from 22% to roughly 1%, with no significant performance tradeoff on Hermes Bench. The model weights, training data and evaluation materials are publicly available so others can test the results independently. The poster cites the release as another example of why access to open model weights matters.
NousResearch Releases Arete-27B, a Qwen3.8-27B Model Post-Trained to Reinforce Western Values
NousResearch announced the release of Arete-27B, described as Qwen3.8-27B post-trained to reinforce Western values. The announcement frames the release against Chinese open models, which the post says are currently the strongest, though the text is truncated. The post positions Arete-27B as a Western-aligned alternative built on a Qwen3.8-27B base model. No further technical details, benchmarks, licensing terms, or availability information were included in the visible text.
Read sourceShow 2 older updatesHide older updates
NousResearch Releases Arete-27B, a Qwen3.8-27B Model Post-Trained to Reinforce Western Values
NousResearch announced the release of Arete-27B, described as Qwen3.8-27B post-trained to reinforce Western values. The announcement frames the release against Chinese open models, which the post says are currently the strongest, though the text is truncated. The post presents the model as a deliberate values-alignment effort built on top of an existing Qwen3.8-27B base. No benchmark results, licensing terms, availability details, or training methodology were included in the visible text. The claim about Chinese open models is an attributed opinion from NousResearch rather than a verified comparison.
Read sourceNousResearch Releases Arete-27B, Fine-Tuned Qwen Model to Reinforce Western Values
NousResearch announced Arete-27B, a post-trained version of Qwen3.8-27B designed to reinforce Western values. The organization states that Chinese open models are currently the strongest available but ship with state-aligned behavior in their weights. To measure this, NousResearch built an 803-prompt evaluation across seven value categories, including freedom of expression, democracy, and privacy, tracking how often a model adopts China's party line. After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models tested, according to the post. Its Hermes Bench score shows no significant performance tradeoffs. The announcement concludes that the values in an open model are one fine-tune away. The evaluation, weights, and training data are available now alongside a full blog post.