Meta released Muse Ember 1.0 on 9 November 2026. Meta calls it the smallest and most efficient model in the Muse family, and aims it at the high-volume, low-difficulty work inside agents: calling tools, routing requests, extracting fields, parsing long documents and handling coding subtasks.
Price and limits
| Prompts up to 200K tokens | Prompts over 200K tokens | |
|---|---|---|
| Input, per 1M tokens | $0.10 | $0.30 |
| Cached input, per 1M tokens | $0.004 | $0.012 |
| Output, per 1M tokens | $0.50 | $1.50 |
- Context: 1M tokens, with up to 512K tokens of output.
- API: supports the Responses API format. The model id is
muse-ember-1.0. - Where: the Meta Model API and Muse Code, and as a preview on Google Cloud.
- Inputs: Google Cloud's listing shows text and images; some reports mention audio and video too, which Meta hasn't confirmed in what we could find.
- Weights: closed, like the rest of the Muse line. Meta hasn't published a parameter count.
How the price compares
At $0.10 / $0.50 it is priced the same as Claude Haiku 5.5 and a little below DeepSeek V4.1 Flash. Its very cheap cached input ($0.004 per million) matters most for agents, which resend the same long context on every step. The price triples for prompts over 200K tokens, so very long sessions cost noticeably more.
| Model | Input / output, per 1M tokens | Context |
|---|---|---|
| Meta Muse Ember 1.0 | $0.10 / $0.50 | 1M |
| Claude Haiku 5.5 | $0.10 / $0.50 | 1M |
| DeepSeek V4.1 Flash | $0.15 / $0.60 | 1M |
| Qwen 3.7 Flash | $0.035 / $0.15 | 1M |