model release

Meta replaces Llama with Muse Spark AI, launches Contemplating mode for complex reasoning

TL;DR

Meta has discontinued its Llama model line and launched Muse Spark as the foundation of its new AI strategy under Meta Superintelligence Labs. The model features a Contemplating mode for complex reasoning tasks and specializes in multimodal perception, health applications, and agentic tasks. Muse Spark is available today in Meta AI apps, with a private API preview for select partners.

2 min read
0

How Muse Spark ranks now

BenchmarkScoreRankSource
Arena Elo1489 Elo18 of 146LMArena, Nov 2026 ↗
GPQA88.4%62 of 165Artificial Analysis, Jul 2026 ↗
SWE-bench Verified74.4%42 of 80Vals AI, Sep 2026 ↗

Rank is against every model in our database with a sourced score on that benchmark, each counted once with its best score.

All Muse Spark specs and scores →

Meta Replaces Llama With Muse Spark AI, Launches Contemplating Mode

Meta has officially discontinued its Llama model line and unveiled Muse Spark as its flagship AI system, signaling a strategic shift in the company's AI division under Meta Superintelligence Labs.

Muse Spark Availability and Positioning

Muse Spark is now available today in Meta AI applications. The company describes it as "the first step on our scaling ladder" for a family of Muse models that will follow. Unlike Anthropic's recent Claude Mythos Preview announcement, which remains in limited testing, Muse Spark has immediate public availability.

Meta positions Muse Spark as a direct competitor to frontier models from other labs. According to Meta, the model offers "competitive performance in multimodal perception, reasoning, health, and agentic tasks." The company acknowledges current performance gaps in long-horizon agentic systems and coding workflows.

Contemplating Mode: Parallel Reasoning Architecture

Muse Spark includes two primary reasoning modes. The base "Thinking" mode benchmarks favorably against Anthropic's Claude Opus 4.6 Max, Google's Gemini 3.1 Pro High, OpenAI's GPT-5.4 Xhigh, and xAI's Grok 4.2 Reasoning models.

Meta's Contemplating mode represents a more significant capability leap, orchestrating multiple agents that reason in parallel. This architecture directly competes with Gemini 3.1 Deep Think and GPT-5.4 Pro. According to Meta, Contemplating mode achieves 58% on Humanity's Last Exam and 38% on FrontierScience Research benchmarks. The company notes this mode will roll out gradually.

Specialized Health Capabilities

Meta has invested significant resources in health-focused reasoning. The company collaborated with over 1,000 physicians to curate training data for medical accuracy. Muse Spark can generate interactive displays explaining health information including nutritional content and exercise physiology.

The model demonstrates multimodal strengths in visual STEM questions, entity recognition, and localization tasks.

API Access and Strategic Timing

Meta is opening a private API preview for select partners alongside the public release. The announcement comes as major AI labs release increasingly capable reasoning models, with Meta positioning Muse as a foundational competitor in this space.

The transition from Llama to Muse represents Meta's pivot toward a unified superintelligence approach under Alexandr Wang's leadership of Meta Superintelligence Labs.

What This Means

Meta's full departure from Llama signals the completion of its shift from open-source model distribution to closed-API commercial deployment. The emphasis on health applications and parallel reasoning agents indicates Meta is targeting enterprise and healthcare verticals beyond general-purpose chat. Contemplating mode's gradual rollout suggests the company is managing compute costs while establishing performance baselines against competitors' reasoning models.

Related Articles

model release

Thinking Machines Lab releases Inkling 2 Mini: 26B-parameter looped MoE with 3B active and 1M-token context

Thinking Machines Lab has released Inkling 2 Mini, a multimodal Mixture-of-Experts model with about 26B physical total parameters and 3B active per token. It repeats its middle layers three times, supports a 1M-token context, and ships with a DSpark drafter. Hugging Face reports it leads 32 of 48 benchmarks against four comparable models.

model release

NVIDIA releases Nemotron-Labs-Sigma: 14 open 3B and 8B continuous diffusion language model checkpoints

NVIDIA released Nemotron-Labs-Sigma, a set of 14 open continuous diffusion language model checkpoints at 3B and 8B parameters. According to MarkTechPost, Sigma denoises text in a learned 16-D embedding space, distills to few-step samplers, and decodes up to 4.20x faster than autoregressive decoding.

model release

Tencent releases Hy4 flagship weights: 770B-parameter MoE, 1,048,576-token context, conditional license

Tencent released downloadable weights for its flagship Hy4 on November 20, with a System Card. The multimodal MoE has a 770-billion-parameter language backbone, 49 billion active parameters per text token and a 1,048,576-token context window, under a license with commercial conditions for model-as-a-service operators.

model release

Resolve AI launches SRE-1, a 27B-parameter model trained to find root causes of production alerts

Resolve AI has launched SRE-1, a 27 billion-parameter model built on open weights and trained with reinforcement learning to find root causes of production alerts. The company also made Primary On-call generally available, putting its AI agents on call ahead of human engineers.

Comments

Loading...