Open-model release wave & tooling (Gemma 4 multimodal, DeepSeek V4, Qwen3.6-Plus/Omni/Plus/Flash/Voxtral/Cohere/Tulu/Arcee Trinity/400B/Chinese landscape/GLM-5.1 754B MoE SOTA agentic, Meta Muse Spark/Avocado OSS + upcoming, Bonsai)
Key Questions
What is GLM-5.1 and its key features?
GLM-5.1 is an open-weight 754B MoE model from Zhipu AI, released on Hugging Face under MIT license. It achieves state-of-the-art on SWE-Bench Pro at 58.4%, supports 200K context length, and sustains 8-hour autonomous execution. It ranks #1 in open source and #3 globally on benchmarks like Terminal-Bench.
Where can I access GLM-5.1?
GLM-5.1 is available on Hugging Face with open weights. It supports local runs, quants, evals, and integrates with tools like HF and urgent locals for deployment.
What are the highlights of Gemma 4?
Gemma 4 is a multimodal model under Apache 2.0 license, available on HF, Ollama, and Workers AI, with 256K context for agentic tasks on edge devices. It offers GPT-5 level performance in vision, math, thinking, running entirely on phone (E2B-27B), as reposted by Demis Hassabis.
What is Meta's Muse Spark?
Muse Spark is Meta's new AI model aimed at scaling towards personal superintelligence, debuting under Alexandr Wang. It gained attention on Hacker News and signals Meta's push into advanced open-source AI.
Tell me about Arcee's open-source model.
Arcee released a 400B-parameter open-source LLM with a small 26-person team on a $20M budget. It's positioned as a major player in the open-source landscape.
What upcoming models is Meta planning to open-source?
Meta plans to release open-source versions of upcoming AI models like Muse Spark, Avocado, and Llama 4 multimodal. Reports confirm this continues their Llama open-source tradition.
What other recent open models were mentioned?
Releases include Qwen3.6-Plus/Code leading OSS Claw, DeepSeek V4, Voxtral, OmniVoice (zero-shot TTS in 600+ languages), and Arcee Trinity. Bindu Reddy teases a major upcoming drop.
What is the status of these open-model releases?
The highlight is in developing status, with urgent focus on HF weights, quants, evals, and timelines for local deployment and tooling.
GLM-5.1 (Zhipu) HF #1 OSS 754B MoE MIT SWE-Bench Pro 58.4%/200K ctx/8hr autonomy; Gemma4 Apache2 HF/Ollama/Workers AI multimodal 256K agentic/edge (E2B-27B vision/math/thinking/GPT-5 lvl phone); Qwen3.6-Plus/Code OSS Claw leads; DeepSeek V4/GLM/Voxtral/OmniVoice; Arcee 400B OSS; Meta Muse Spark/Avocado Llama4 multimodal OSS incoming; Bindu teases major drop. Urgent locals/HF weights/quants/evals/timelines.