1/5 Meta is back in the AI model API game. Muse Spark 1.1 is a new multimodal reasoning model for agentic tasks, now available through the Meta Model API public preview. ai.meta.com/blog/introdu...
Introducing Muse Spark 1.1 ai.meta.com
AI Weekly's analysis
→
- The Meta Model API natively supports both OpenAI Chat Completions and Anthropic Messages formats, removing migration cost for developers already on rival APIs.
- Muse Spark 1.1 leads MCP Atlas tool-use (88.1) but trails GPT-5.5 on DeepSWE 1.1 (53.3 vs 67.0), placing it as an orchestration model.
- Zuckerberg broke a three-year X silence to announce the launch, a move multiple outlets flagged as a deliberate platform-level strategic signal.
Read full analysis →
Meta Muse Glimmer 👇🏼 Meta's 30B open-weight agentic model targets local use. It supports tool use, failure recovery, text+image input, adjustable reasoning, and 100+ languages. License: Apache 2.0. research.meta.ai/blog/introdu... #ai #llm
Introducing Muse Glimmer: An Open Agentic Model That Runs on Your Device research.meta.ai
AI Weekly's analysis
→
Read full analysis →
Interesting… techcrunch.com/2026/08/26/n...
Nvidia closes in on Hugging Face acquisition | TechCrunch techcrunch.com
AI Weekly's analysis
→
Read full analysis →
5/5 Muse Spark 1.1 is also available in Thinking mode in the Meta AI app and on meta.ai. The bigger signal: Meta is putting an agent-oriented model behind a developer API, putting it in direct competition with Anthropic, OpenAI, and Google.
meta.ai
Started to work on a sequence of tutorials focusing on running LLMs locally on Apple Silicon with MLX LM 👇🏼 github.com/RamiKrispin/... #ai #llm
GitHub - RamiKrispin/local-llms-with-mlx: Running LLMs locally with MLX LM github.com
Tau is a nice open-source project from Hugging Face for learning how coding agents are built. It separates the stack into tau_ai, tau_agent, and tau_coding: provider stream, reusable agent harness, and terminal coding app. Repo: github.com/huggingface/... #ai #aiagents #openso…
GitHub - huggingface/tau: a minimalist agent that teaches you to create coding agents github.com
A new tutorial in my Docker 101 series github.com/RamiKrispin/...
GitHub - RamiKrispin/docker-for-ai github.com
MiniMax-M3 is a open-weighted native multimodal model with 1M context, ~428B total and ~23B active parameters. It handles text, images, and video, uses MiniMax Sparse Attention, and supports local serving with SGLang, vLLM, and Transformers. Repo: github.com/MiniMax-AI/M... #a…
GitHub - MiniMax-AI/MiniMax-M3 github.com
OpenScience is a new open-source AI workbench for scientific research: browser workspace + research agents that can read literature, write/run code, run experiments, query scientific databases, and write up results. github.com/synthetic-sc... #ai #opensource
GitHub - synthetic-sciences/openscience: The open-source AI workbench for scientific research github.com
DiffusionGemma — Google's new open text-diffusion model. 26B MoE that drafts 256 tokens in parallel instead of one at a time. Up to 4x faster on local GPUs. Bi-directional attention, Apache 2.0. huggingface.co/google/diffu... #gemma #ai
google/diffusiongemma-26B-A4B-it · Hugging Face huggingface.co
Here is how you can run GLM 5.2 locally, if you simply have a machine with 256GB to 810GB RAM 😅 unsloth.ai/docs/models/... #ai #llm
GLM-5.2 - How to Run Locally | Unsloth Documentation unsloth.ai
AI Weekly's analysis
→
Read full analysis →
Unsloth released new GLM settings that enable running the model on a 128 Gb RAM machine 3.3x faster using optimized decoding 👇🏼 unsloth.ai/docs/models/...
GLM-5.3-Flash: How to Run Locally | Unsloth Documentation unsloth.ai
AI Weekly's analysis
→
- GLM-5.3-Flash is Z.ai's 320B-parameter (18B active) multimodal open model with a 1,048,576-token maximum context.
- Unsloth's dynamic 1-bit quant weighs 93GB and reportedly retains 71% of top-1% accuracy versus the 642GB BF16 build.
- A September 4 update adds Multi-Token Prediction, lifting B200 throughput from 58.6 to 86.5 tokens per second at 4,096 context.
Read full analysis →
