Releases·Model Release
TheDrummer releases Artemis 31B v1 and v1.1 finetunes, teases crowdsourced inference platform 'BeaverNet'
Community model creator TheDrummer released Artemis-31B v1 and v1.1 on Hugging Face, marking a return after a months-long hiatus. The 31B models target prose and writing quality, with v1.1 fixing quirks like stuttering found in v1. The release builds on the recent wave of Gemma-based fine-tunes the creator credits with ending a year-long drought of new base models.
178 engagement1 sourceFri, Sep 4, 2026, 03:18 PM
AI Overview
- TheDrummer releases Artemis 31B v1 and v1.1 on Hugging Face: v1 excels at prose but has quirks like stuttering; v1.1 prioritizes stability
- TheDrummer teases a HordeAI-like crowdsourced inference platform, nicknamed 'BeaverNet', where local hosters share compute for free
- TheDrummer's backlog includes finetunes of Gemma E2B/E4B/12B/26BA4B, Qwen 3.8 27B, Muse Glimmer 30B, and Mistral Medium 3.5 128B
- TheDrummer credits Skyfall 31B v4.2, Rocinante 12B X/16B XL, Anubis 70B v1.2, Valkyrie 49B v2.1, and Anubis Mini 8B v1 as recent community-liked finetunes
Related
Releases·Model Releasefeatured
Ornith AI releases Ornith 1.5 family: 9B, 35B-A3B, and 397B models
Ornith AI released the Ornith 1.5 family of open-source LLMs, spanning 9B dense, 35B MoE, and 397B MoE, trained with self-improving strategies. The models achieve state-of-the-art performance among open-source models of comparable size, with the 397B matching Claude Opus 4.8 on reasoning, agentic, and coding tasks. Community quantizations are already available for the smaller models.
Ornith AI released the Ornith 1.5 family of open-source LLMs, spanning 9B dense, 35B MoE, and 397B MoE, trained with self-improving strategies. The models achieve state-of-the-art performance among open-source models of comparable size, with the 397B matching Claude Opus 4.8 on reasoning, agentic, and coding tasks. Community quantizations are already available for the smaller models.
2101.0k engagement3 sourcesredditWed, Aug 19, 2026, 01:53 PM
Releases·Tool Release
SGLang v0.5.18 adds Muse Glimmer, Intern-S2-Mobius, SANA-Video support
SGLang v0.5.18 is released with support for new models including Muse Glimmer, Intern-S2-Mobius, and SANA-Video. The release includes 710 PRs from 212 contributors, expanding the inference engine's model coverage.
SGLang v0.5.18 is released with support for new models including Muse Glimmer, Intern-S2-Mobius, and SANA-Video. The release includes 710 PRs from 212 contributors, expanding the inference engine's model coverage.
0 engagement1 sourcegithubSat, Aug 22, 2026, 12:09 AM
Social·Community
AI News roundup: Seedance 2.5 and other model releases
A YouTube video from Future Tools summarizes the week's AI news, highlighting the release of Seedance 2.5 and other new models. The video is a roundup of notable developments for practitioners.
A YouTube video from Future Tools summarizes the week's AI news, highlighting the release of Seedance 2.5 and other new models. The video is a roundup of notable developments for practitioners.
2.3k engagement1 sourceyoutubeFri, Aug 14, 2026, 03:00 PM
Releases·Model Release
OpenBMB releases MiniCPM5-2B, claims best among 4B models
OpenBMB released MiniCPM5-2B, a 2-billion-parameter model that they claim is the best locally ranked AI model among 4B-parameter models. The model is not yet available on Hugging Face as of the post date.
OpenBMB released MiniCPM5-2B, a 2-billion-parameter model that they claim is the best locally ranked AI model among 4B-parameter models. The model is not yet available on Hugging Face as of the post date.
240 engagement1 sourceredditMon, Jul 20, 2026, 01:47 PM
Social·Community
Community impressions of Nanbeige 4.2 3B with DSpark on Hugging Face
A Reddit user highlights Nanbeige4.2-3B-DSpark, a small model on Hugging Face, claiming it outperforms Qwen 3.5 9B. They report the DSpark variant fixed the model's slow autoregressive speed, reaching roughly 35 tokens/s. The post targets users with limited GPU resources.
A Reddit user highlights Nanbeige4.2-3B-DSpark, a small model on Hugging Face, claiming it outperforms Qwen 3.5 9B. They report the DSpark variant fixed the model's slow autoregressive speed, reaching roughly 35 tokens/s. The post targets users with limited GPU resources.
47 engagement1 sourceredditMon, Aug 31, 2026, 07:08 PM