OpenAI expanded availability of its GPT-5.6 family—paid users now get GPT-5.6-Sol by default while free users have unlimited access to GPT-5.6-Luna—and rolled out GPT-5.6-cyber as part of its Daybreak cybersecurity program.
Meanwhile, Meta released Muse Glimmer, an open-weight 30B-parameter model for agentic tasks, and committed to releasing Muse Spark 1.2, its best-performing model.
The moves underscore competition between closed and open model strategies.
What happened
OpenAI made GPT-5.6-Sol the default for paid ChatGPT users and GPT-5.6-Luna available for free users with unlimited chats. The company also expanded its Daybreak cybersecurity program and released GPT-5.6-cyber alongside it. Meta released Muse Glimmer, an open-weight model with 30B parameters tuned for agentic tasks including computer use, and announced plans to release weights for Muse Spark 1.2, its best-performing model to date.
Why it matters
The broader rollout of GPT-5.6 variants gives more users access to OpenAI's latest reasoning capabilities—paid users get a stronger default, while free users gain unlimited access. Meta's open-sourcing of Muse Glimmer and upcoming Muse Spark 1.2 reinforces the shift toward open-weight models that developers can run themselves, offering an alternative to closed commercial offerings.
What to watch
OpenAI's next mainstream model, Astra, is positioned as critical for cybersecurity but has not yet been released. Bloomberg also reported OpenAI is developing a smart speaker without a display in a doughnut-like shape, priced over $300, arriving in 2027.
OpenAI updated its standard ChatGPT product lineup to prioritize its newest reasoning models. Paid subscribers now receive GPT-5.6-Sol as their default model, while free users gain unlimited chats powered by GPT-5.6-Luna, which operates at lower reasoning levels. The company also expanded its Daybreak cybersecurity program—a defensive initiative aimed at businesses and enterprises—and bundled it with a new model, GPT-5.6-cyber, specifically optimized for security tasks. OpenAI's next major mainstream release, Astra, is expected to be especially important for cybersecurity applications, but the company has not yet set a release timeline, signaling a longer development or validation cycle.
Meta took the opposite path by intensifying its open-source strategy. The company released Muse Glimmer, an open-weight model with 30 billion parameters, designed and tuned for agentic tasks—including computer use, where AI systems interact with digital interfaces as a human would. Meta also announced it will soon release the weights of Muse Spark 1.2, which the company describes as its best-performing model to date. This dual release underscores Meta's philosophy that open-source models can compete on capability while building developer trust. CEO Mark Zuckerberg reinforced this vision in a 6,000-word statement on Meta's AI philosophy, emphasizing three pillars: personal agents that understand users and their goals, tools for creativity, business, and education, and free or affordable access to these capabilities.
The two announcements reflect a widening strategic divergence. OpenAI is layering its models by reasoning capability and use case (consumer chat, paid tiers, cybersecurity-focused variants), betting that premium and specialized models justify higher pricing. Meta is betting that open weights—accessible to any developer—will dominate the ecosystem over time, with Muse Spark 1.2 positioned as a flagship capable of competing with closed systems.
OpenAI and Meta are pursuing divergent strategies in large language models. OpenAI is consolidating its consumer offerings around the GPT-5.6 family, segmenting access by subscription tier and extending its cybersecurity focus through GPT-5.6-cyber and the Daybreak program. Meta, by contrast, is doubling down on open-sourcing—releasing Muse Glimmer with 30B parameters and committing to release Muse Spark 1.2, its flagship model. This mirrors a broader industry split: closed, frontier models (OpenAI's approach) versus open weights that developers can self-host (Meta's approach). The timing matters because it signals Meta's renewed commitment to open-sourcing after a period of focus on proprietary models. Both strategies reflect the tension in the AI market between control and accessibility: OpenAI maintains tighter grip on its most capable systems while Meta bets on developer adoption and trust through transparency.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
The AI news that matters, in one minute each morning.
Sign up free