Meta Superintelligence Labs' multimodal, agentic reasoning model — Meta's first paid API tier, now on its 1.2 release.
Muse Spark 1.2 runs $1.25 / 1M input and $4.25 / 1M output on the Meta Model API — the same standard-tier rate as 1.1 — with cached input now disclosed at $0.15 / 1M, plus $20 in free credits to start.
Meta's official model catalog now lists Muse Spark 1.2 as the current release, superseding Muse Spark 1.1 (introduced July 9, 2026 as Meta's first-ever paid model). Standard-tier pricing carries over unchanged at $1.25 input / $4.25 output per 1M tokens, with $20 in free credits to start. Meta's docs also newly disclose a cached-input rate of $0.15 per 1M, and a separate "Contributor" tier ($0.10/$0.20 per 1M) for developers who opt in to having their data used for training.
Muse Spark remains a multimodal reasoning model built for agentic tasks, with strength in tool and computer use, coding, and multimodal understanding. It can actively manage its 1M-token context window rather than treating it as a static buffer.
Muse Spark 1.2 can write and debug code, use software and external tools, understand text, images, and video, and carry out complex multi-step tasks with less human intervention. That agentic profile — plus a low output price — positions it against the mid-tier flagships from OpenAI, Anthropic, and Google, at a time when Meta is accelerating its AI push.
| Model | Input / 1M | Output / 1M | Context |
|---|---|---|---|
| Meta Muse Spark 1.2 | $1.25 | $4.25 | 1M |
| GPT-5.6 Terra | $2 | $12 | 1M |
| Claude Sonnet 5 | $2 | $10 | 1M |
| Gemini 3.1 Pro | $2 | $12 | 1M |
| Grok 4.6 | $2 | $6 | 500K |
On price, Muse Spark 1.2 undercuts every mid-tier flagship listed here on output tokens — cheaper than Gemini 3.1 Pro, Claude Sonnet 5, and Grok 4.6 — while claiming benchmark wins over Gemini. It remains an aggressive price point for Meta's paid model line.