Muse Spark 1 1 Prompting

# Meta Muse Spark 1.1 Prompting Guide

Checked: 2026-07-12

Muse Spark 1.1 is Meta's **multimodal reasoning model** for agentic work,
computer use, coding, and multimodal understanding. It appears in Thinking mode
in Meta AI and through the Meta Model API public preview. The model name alone
does **not** grant a repository editor, browser, or computer-control harness;
the client supplies tools.

| Property | Value |
| --- | --- |
| Role | Lower-cost fast multimodal reasoning |
| Independent snapshot | ~51 Intelligence Index at xhigh; ~116 tok/s; 1M context; ~$1.25/$4.25 per 1M |
| Surfaces | Meta AI Thinking mode; Meta Model API public preview |
| Equal-score warning | Rounded Intelligence near Luna Max does not imply equal Codex coding reliability |

## When to Choose Muse Spark 1.1

Choose Muse when:

- you need multimodal understanding plus reasoning at lower cost;
- throughput matters;
- your own client provides tools (browser, computer use, code runner).

Prefer Luna/Terra/Grok/Sol when:

- you need a measured Codex coding-agent configuration;
- you need GPT-5.6 product integration (Work/Codex);
- GUI computer-use quality is unproven for your app.

## Prompting Principles

1. **Name the tools** the client actually exposes.
2. **Separate modalities** in the prompt (image facts vs text instructions).
3. **Keep one goal** per turn for agent loops.
4. **Require evidence** (screenshots, DOM snippets, file paths, command output).
5. **Do not assume** hidden computer control.

## Effort / Thinking Modes

Exact API enum names can change; independent testing has used an **xhigh-like**
deep setting. Treat Thinking vs non-Thinking (Meta AI) and API reasoning levels
as compute dials, not different model families.

| Mode band | Use | Prompt emphasis |
| --- | --- | --- |
| Fast / low thinking | Extraction, captioning, simple Q&A on clear inputs | Schema, short answers |
| Default thinking | Multistep tool use, coding with clear tests | Goal, tools, checks |
| Deep / xhigh | Ambiguous multimodal problems, multi-app computer tasks | Hypotheses, checkpoints, stop rules |

### Fast template