Muse Spark 1.2

Meta released 2026-08-05 proprietary

$2.00 per million tokens, balanced

Nothing is both better and cheaper.

Under a balanced workload, no other model in the catalogue scores higher and costs less. This model is on the value frontier.

Our take

editorial — not a measurement

Meta's current frontier entry and a real one — LMArena 1498 at xHigh effort puts it fourth overall, above every OpenAI model on that board. At $1.25/$4.25 with a 1M window and the broadest modality set in the dataset (text, image, video, file, audio input) it is aggressively priced for the position. The catch is that it appears to be proprietary despite Meta's open-weight history: no Hugging Face repo is listed and no first-party Meta API pricing page could be found, so the only price signal is the OpenRouter listing. A separate 'contributor' tier exists at $0.10/$0.20.

Strengths

  • LMArena 1498 at xHigh — 4th overall, above all OpenAI entries
  • $1.25/$4.25 for frontier-class placement
  • Widest input modality set here: text, image, video, file, audio
  • 1M context window

Weaknesses

  • No first-party Meta pricing page located — OpenRouter is the only source
  • No Hugging Face repo; appears proprietary despite Meta's Llama history
  • No AA intelligence score captured
  • Max output tokens not published

Reach for it when

  • Multimodal frontier work at mid-tier prices
  • Audio and video input pipelines

Avoid it if

  • You require first-party vendor pricing before adopting
  • You need open weights from Meta — use Llama 4 or Muse Glimmer

Sources: openrouter.aiarena.ai

Every price dimension

Input$1.25/M
Output$4.25/M
Cached input$0.150/M
Web search$0.0025/call

Independent scores

Intelligence56.8
Coding72.2
Agentic49.3

Sources are listed separately rather than averaged. Across the 62 models both have scored they correlate at r = 0.806 — close agreement overall, but four models rank very differently between them, and a blended score would hide exactly those.

Best rankings by task

  • website #2of 151 1336
  • dataviz #2of 142 1370
  • codecategories #3of 145 1344
  • gamedev #4of 144 1352
  • uicomponent #4of 140 1352
  • 3d #6of 136 1348
  • webapps #9of 56 1264
  • androidnative #10of 54 1247

Capability

Context window1.0M
Input modestext, image, video, file, audio
Tool useyes
Reasoningalways on
Open weightsno

Provenance

Price sourceopenrouter.ai
Fetched2026-08-24
Quality dataverified
Cross-checkedvendor page

Muse Spark 1.2 is a reasoning model from Meta, designed for complex agentic tasks. It accepts text, images, video, audio, and PDF documents, returns text, and offers a 1M-token context...

Markdown for LLMs