Disney and Adobe agree to develop custom AI models
The multiyear partnership will initially focus on creative development at Disney Experiences and StudioLAB, with additional work planned for ESPN.
Meta launches Muse Ember for high-volume AI agent tasks
Input pricing is 10 cents for every million tokens, with Meta positioning the model to carry out repetitive work alongside its larger Muse Wonder model.
Meta Superintelligence Lab introduced Muse Ember on November 9, 2026, targeting repeated tasks performed by AI agents. Meta described Ember as its smallest and most cost-efficient model so far, with input pricing of 10 cents for every million tokens.
The model is intended to handle tool use, smaller coding jobs, information extraction, routing and processing lengthy documents. Meta envisions pairing it with a larger model such as Muse Wonder, which would plan the work while Ember carries out tasks at volume.
For customer support, that arrangement could involve a larger model setting rules for an issue and a lighter model applying them across tens of thousands of cases.
Meta used a composite score from Artificial Analysis to present Ember as less expensive and more capable than Anthropic’s Haiku and OpenAI’s Luna. The company also said Ember can process multiple media formats, including text, images and audio, without requiring a separate model.
Ember brings the Muse model lineup to four. Spark arrived in April, followed by its open source counterpart Glimmer in August and the larger Wonder model in October.