OpenAI's Dots vs Meta's Muse: Agents Now Compete With Free
At DevDay, OpenAI introduced Dots, an agent built on GPT-6 Astra, aimed squarely at Meta's Muse platform. For builders, the real question is what a hosted agent can offer against open-weight models that run locally.
By TRAGenX Desk
OpenAI used its annual DevDay to launch Dots, an agent it pitched as "real-deal AI" and which runs on GPT-6 Astra. The Verge reports that the launch is aimed mainly at Meta, whose Muse agent platform has had a strong early run. The Verge's headline asks whether OpenAI can compete with free.
What actually happened
The announcement is a platform move, not just a model release. Both companies now sell the same thing: an agent that acts on your behalf, rather than a chatbot you query. The competition has moved from benchmark scores to who owns the agent layer.
We could only confirm a few details from the primary coverage: the name, the underlying model and the competitive framing. We are leaving out pricing, availability and feature claims we could not verify.
Why "free" is the part that matters
Meta has pushed hard on openness. Its Muse Glimmer model is described as a small, fully open model that runs AI agents entirely on-device. The Rundown reports that it beats similar-sized rivals such as Gemma4 and Qwen3.6 on a range of agentic, coding and reasoning tests.
Treat that as the vendor-adjacent claim it is, and run your own evals. Even so, the direction is clear. When a capable agent model runs on a laptop with open weights, a hosted frontier agent has to justify its price with something other than raw capability.
What builders should take from it
- Don't marry a model. Put the agent behind an interface you control so you can swap a hosted model for a local one, or the reverse, when the economics change.
- Own your evals. Vendor launch claims are not a substitute for a regression suite built from your own tasks, including the failures.
- Permissions beat prompts. An always-acting agent needs scoped credentials, explicit approval for irreversible actions and audit logs. This matters most in trading and fintech, where an agent action moves money.
- Match the deployment to the data. On-device models suit sensitive data and offline workflows, while hosted models may suit heavier long-horizon work. Test both rather than assuming.
Our read
Agent platforms are becoming a commodity layer, and the differentiation is moving to reliability and control. The teams that do well will be the ones who can show their agent behaves under constraint, whichever model sits underneath.
FAQ
Frequently asked questions
- What is OpenAI Dots?
- According to The Verge, Dots is an AI agent OpenAI announced at DevDay, powered by GPT-6 Astra and positioned as a rival to Meta's Muse agent platform.
- What is Meta Muse Glimmer?
- The Rundown describes it as a small, fully open-weight Meta model that runs AI agents entirely on-device, and says it outperforms similar-sized models such as Gemma4 and Qwen3.6 on agentic, coding and reasoning tests.
- Should I build on a hosted agent or an open-weight one?
- Design so you don't have to choose permanently. Abstract the model behind your own interface, benchmark both on your real tasks, and weigh cost, data sensitivity and the reliability you can measure.
Sources
- OpenAI's new agent is a shot at Meta — but can it compete with free? — The Verge AI
- Meta returns to its open source roots — The Rundown AI