Meta Launches Muse Ember 1.0, Its Smallest and Cheapest Model Built for Agent Workloads

Meta released Muse Ember 1.0 on November 10, 2026 — its smallest and cheapest model, priced at $0.10 per million input tokens and $0.50 per million output tokens, targeting the high-volume agent tier where Claude Haiku and GPT-6 Luna currently compete. The model handles text, images, audio, and video natively and supports a 1-million-token context window.

This is Meta's first credible entry into the market for cheap, fast models purpose-built to run inside AI agent pipelines.

What Meta Released

Muse Ember 1.0 is available immediately through the Meta Model API and inside Muse Code, Meta's AI coding agent. Key specifications:

• Context window: 1 million tokens (roughly 750,000 words, or a full novel plus reference material)
• Modalities: text, images, audio, and video — natively, not via plug-in tools
• Pricing: $0.10/M input tokens, $0.50/M output tokens
• Availability: Meta Model API and Muse Code (day one)

Meta claims Ember delivers higher intelligence-per-dollar than competing small models. The company did not publish head-to-head benchmark comparisons at launch.

The Market It's Entering

The small, cheap, fast model segment is where the AI industry's volume economics actually live. Building a customer service agent, a document processing pipeline, or an automated coding assistant means making millions of API calls per day. At that scale, the difference between $0.10/M and $0.25/M input tokens is a significant operational cost decision.

The current market leaders in this tier:

• Claude Haiku — Anthropic's budget model, cut to competitive pricing in October 2026
• GPT-6 Luna — OpenAI's small-model entry
• Gemini Flash — Google's cost-optimized tier

Ember's native multimodality is a meaningful differentiator. Most competing small models handle images adequately but treat audio and video as afterthoughts, often routing them through separate systems. If Ember genuinely processes all four modalities in a single pass at this price point, it simplifies the architecture of agent pipelines that need to handle mixed media.

Why Meta Is Here Now

Meta's large-model story — the Muse flagship and Llama open-weight releases — has been competitive. Its small-model story has been weaker. Ember is a direct response to developer feedback: the Muse ecosystem needed a cheap, capable model that developers could use for the high-volume, low-cost steps of an agent workflow without leaving Meta's API.

The 1-million-token context window is aggressive for a small model. Most competitors cap their budget tier at 128,000 to 200,000 tokens. Longer context matters for tasks like processing a full codebase, an entire legal document set, or hours of audio transcription in a single call.

What to Watch

Meta has not published third-party benchmark results for Ember yet. Independent evaluations from organizations like LMSYS or Artificial Analysis will determine whether the intelligence-per-dollar claim holds up. Watch for Anthropic and OpenAI to respond with pricing adjustments or capability announcements in the small-model tier — Ember's launch puts direct pressure on margins at the bottom of their pricing stack.