GPT-56T 861 —
MUSE-SPK 837 —
GPT-56SC 789 -0.1%
GLM-5 781 —
CL-OP55X 779 -0.1%
GROK-46H 779 -0.1%
QWEN-38X 748 —
GPT-6A 743 —
KIMI-K3X 742 —
CL-FAB5H 697 -0.1%
CL-OP5H 674 -0.1%
GEM-38FH 672 —
CL-OP5X 669 -0.1%
CL-OP55H 667 -0.1%
CL-OP46H 656 -0.2%
CL-OP47H 647 -0.2%
GPT-56S 617 -0.2%
GEM-37FH 609 -0.2%
GEM-36FH 592 -0.2%
CL-OP48H 587 -0.2%
CL-OP47 580 -0.2%
GEM-35FH 579 -0.2%
GPT-55H 540 -0.2%
INKL 531 —
GEM-31P 511 -0.2%
CL-OP46 498 —
GEM-3P 498 —
CL-OP48 492 —
GPT-52 464 —
GPT-55 423 —
GPT-56T 861 —
MUSE-SPK 837 —
GPT-56SC 789 -0.1%
GLM-5 781 —
CL-OP55X 779 -0.1%
GROK-46H 779 -0.1%
QWEN-38X 748 —
GPT-6A 743 —
KIMI-K3X 742 —
CL-FAB5H 697 -0.1%
CL-OP5H 674 -0.1%
GEM-38FH 672 —
CL-OP5X 669 -0.1%
CL-OP55H 667 -0.1%
CL-OP46H 656 -0.2%
CL-OP47H 647 -0.2%
GPT-56S 617 -0.2%
GEM-37FH 609 -0.2%
GEM-36FH 592 -0.2%
CL-OP48H 587 -0.2%
CL-OP47 580 -0.2%
GEM-35FH 579 -0.2%
GPT-55H 540 -0.2%
INKL 531 —
GEM-31P 511 -0.2%
CL-OP46 498 —
GEM-3P 498 —
CL-OP48 492 —
GPT-52 464 —
GPT-55 423 —
← Back to feed

Meta Enters the Frontier: Muse Spark Debuts at #4 on the Intelligence Index

Meta has re-entered the frontier AI race. On April 8, Meta Superintelligence Labs — the new division formed after Zuckerberg’s multi-billion dollar overhaul of Meta’s AI efforts — launched Muse Spark, the first model in a new family aimed at what Meta calls “personal superintelligence.”

It is the first frontier-class model from Meta since Llama 4 Maverick in April 2025, and a significant strategic break from Meta’s open-source tradition: Muse Spark is proprietary, closed-weights, and not available for local deployment.

Key Numbers

  • Intelligence Index (Artificial Analysis): 52 — 4th overall, behind Gemini 3.1 Pro (57), GPT-5.4 (57), Claude Opus 4.6 (53)
  • Humanity’s Last Exam: 39.9% — trails only Gemini 3.1 Pro (44.7%) and GPT-5.4 xhigh (41.6%)
  • Contemplating mode (HLE): 58% — Meta’s multi-agent reasoning stack, competing with GPT Pro and Gemini Deep Think
  • Tau2-bench Telecom: 92% — on par with frontier peers
  • Token efficiency: 58M output tokens to run the full Intelligence Index, compared to 157M for Claude Opus 4.6 and 120M for GPT-5.4 xhigh
  • Arena ELO: ~1493 — top 5 on Chatbot Arena

What It Is

Muse Spark is a natively multimodal reasoning model supporting text and image input, tool use, visual chain-of-thought, and multi-agent orchestration. It is the first product of a ground-up rebuild of Meta’s training stack — the company claims it reaches the same capability level as previous models with over 10× less compute than Llama 4 Maverick.

Contemplating mode is Muse Spark’s answer to extended thinking. It orchestrates multiple sub-agents reasoning in parallel before returning a response. Meta says this closes the gap with frontier reasoning modes from Google and OpenAI on hard tasks.

Where It Falls Short

Coding is the notable weakness. On Terminal-Bench Hard, Muse Spark trails Claude Sonnet 4.6, GPT-5.4, and Gemini 3.1 Pro. Long-horizon agentic tasks are a stated gap Meta acknowledges in their release post.

Access and Pricing

Muse Spark is available free at meta.ai and the Meta AI app, with a gradual rollout to WhatsApp, Instagram, Facebook, and Messenger. A private API preview is open to select partners — no public pricing or documentation has been published yet. No public API means Stack Futures currently imputes its cost index from the dataset median.

Meta says open-sourcing future models is the plan, but no timeline has been given for Muse Spark specifically.

What It Means

This is the most significant move Meta has made in frontier AI. Llama 4 was a disappointment — Muse Spark is not. Jumping from an Intelligence Index score of 18 (Llama 4 Maverick) to 52 in a single release is a leap that puts Meta back in the conversation with Anthropic, Google, and OpenAI.

The decision to go closed-weights is a signal that Meta is playing a different game now — building a product, not a research commons. With larger Muse models already in development and a distribution network of 3 billion users across its apps, Meta’s path to scale is unlike any other lab in the field.