Meta launches Muse Spark Open model
Meta releases Muse Spark Open, a new ‘open-weight’ AI model
Saturday, 17 October 2026
Photo by Haseeb Modi on Unsplash
Warm-up
- What AI tools have you used, if any?
- Should powerful AI be open or closed? Why?
- Where could AI help you study or work?
Vocabulary
- launch
- to start or release a new product or service
- rival
- to compete with someone or something of similar strength
- outperform
- to do better than another person, product or group
- parameter
- a value in a system that controls how it behaves
- modify
- to change something in order to improve or adjust it
- safeguard
- a rule or tool designed to prevent harm or misuse
- optimise
- to make something work as well and efficiently as possible
- partnership
- a working relationship between two organisations
Reading
Listen to this story
British EnglishMeta has launched Muse Spark Open, an open‑weight AI model it says can rival, and sometimes outperform, leading Chinese systems. The company says it excels at multi‑step reasoning, complex software engineering, and is built for coding and agentic work. Unlike closed models such as Anthropic's Claude or OpenAI's ChatGPT, developers can download the model's parameters (the weights and biases) and modify them.
Chinese developers have been leading the open‑weight field in recent years, with Alibaba's Qwen, Moonshot AI's Kimi K3 and Z.ai's GLM seen as frontrunners. Meta argues its model now competes on a similar level. It also says "strong safeguards" are integrated to reduce misuse.
Muse Spark Open is optimised to run on Apple's new M5 Ultra chips, signalling an unusual partnership between the two firms. The model requires significant computing power: according to Meta, users can run it on two Mac Studios with M5 Ultra chips to create a desktop "supercomputer".
The launch comes days after Meta announced a new closed model that it says rivals the most advanced versions of ChatGPT and Claude. Meta had been focusing on open models, but it is now pushing to compete more directly with OpenAI and Anthropic across both approaches.
Comprehension
- What kind of AI model is Muse Spark Open?
- How is an open‑weight model different from a closed one?
- Which Chinese models are named as frontrunners?
- What hardware is the model optimised to run on?
- What unusual desktop setup did Meta describe for running the model?
Grammar focus
Past perfect continuous
Use had been + verb‑ing for an action in progress before a later past point. Common cues: before, by the time, for/since. Form: had been + -ing. (Example 2 constructed.)
- Meta had been focusing on open models, but it is now pushing.
- For months before release, teams had been prototyping tools.
Grammar exercise
Complete the sentences with the past perfect continuous.
- By the launch, reviewers (compare) ___ systems for months.
- Until the release, developers (fine-tune) ___ safety prompts for days.
- By the time chips shipped, teams (configure) ___ rigs for weeks.
- Before adoption, many labs (download) ___ weights for hours.
- By then, moderators (review) ___ safeguards for weeks.
Discussion
- Should powerful AI models be open‑weight? Give reasons.
- Does Meta’s partnership with Apple surprise you? Why or why not?
- Do strong safeguards make you trust AI more? Explain.
- Is high computing power a fair barrier to using advanced AI?
Vocabulary, reading, and exercises with instant feedback
Spotted a mistake in this lesson? Tell us.