Muse Spark

Overview

Muse Spark is the first model from Meta Superintelligence Labs (MSL), a natively multimodal reasoning model with built-in tool use, visual chain of thought, and multi-agent orchestration. Built over nine months from a ground-up rebuild of Meta's AI stack, it powers the Meta AI assistant across meta.ai, WhatsApp, Instagram, Facebook, and Messenger.

The model is intentionally compact and fast, yet scores 52 on the Artificial Analysis Intelligence Index v4.0 — reported at 52, though benchmark methodology and comparability across models should be interpreted cautiously. Its standout feature is Contemplating mode, a multi-agent parallel reasoning system that reported a score of 50.2 on Humanity's Last Exam, though independent verification and benchmark conditions are not fully disclosed. Muse Spark is currently free to use through Meta's platforms, though Meta has not published detailed usage limits or long-term pricing guarantees.

Key Features

• Contemplating mode — Spawns and coordinates multiple AI agents in parallel to solve complex problems, reportedly achieving 58% on Humanity's Last Exam with tools and 38% on FrontierScience Research benchmarks (not independently verified), competing with Gemini Deep Think and GPT Pro.

• Native multimodal perception — Sees and understands images directly rather than relying on text descriptions. Snap a photo of a shelf and Muse Spark can identify items, rank nutritional content, or compare products visually.

• Visual coding — Creates custom websites, interactive dashboards, and retro arcade games directly from prompts, generating shareable experiences without requiring development skills.

• Health-informed responses — Developed with reported input from over 1,000 physicians to improve health-related explanations, though the scope of involvement is not fully detailed. The model can generate interactive displays that explain health information such as nutritional content, muscle activation during exercise, and medical imagery.

• Multi-agent orchestration — Launches parallel subagents for complex tasks. For trip planning, one agent drafts itineraries, another compares locations, and a third identifies family-friendly activities — all simultaneously.

• Context-aware community integration — Can surface public posts, discussions, and recommendations within Meta platforms, depending on availability and region. It pulls from trending community discussions and creator-curated shopping content directly within conversations across Meta platforms.

How It Compares

Muse Spark enters a competitive AI chatbot landscape as Meta's first closed-source model:

• vs. GPT-5.4: GPT-5.4 scores higher on the Intelligence Index (57 vs. 52) and offers a mature public API. Muse Spark's advantage is free unlimited access through Meta platforms and deeper social media context integration.
• vs. Gemini 3.1 Pro: Gemini matches GPT-5.4 at 57 on the index and offers extensive Google ecosystem integration. Muse Spark competes on Contemplating mode's parallel reasoning and its native presence across WhatsApp, Instagram, and Facebook.
• vs. Claude Opus 4.6: Claude scores 53, slightly above Muse Spark, and excels at long-context reasoning and coding. Muse Spark differentiates through free access, multimodal visual perception, and social platform integration.

Pricing & Plans

Muse Spark is completely free to use:

• Meta AI (free): Full access through meta.ai, the Meta AI app, WhatsApp, Instagram, Facebook, and Messenger. No subscription fees are currently required, and Meta has not publicly disclosed usage limits.
• API access: Currently limited to a private preview for select partners; no public API pricing or release timeline has been officially announced.
• Contemplating mode: Rolling out gradually to users at no additional cost.

Meta has not announced paid tiers, and future monetization details remain unclear.

Best For

• Everyday users who want a free, capable AI assistant without paying for ChatGPT Plus or Claude Pro
• WhatsApp and Instagram users who prefer AI assistance within apps they already use daily
• Users seeking multimodal help — identifying products from photos, comparing items visually, or getting health information from images
• Students and researchers who need strong reasoning capabilities at no cost, particularly through Contemplating mode
• Creative users who want to generate interactive websites, dashboards, or mini-games from text prompts

FAQ