DevOps / SRE / Platform · 10.08.2026, 15:10 UTC
Meta Muse Code vs. Fable 5: Meta Muse is cheaper, but at what cost?
| Schweregrad | info |
|---|---|
| Kategorie | DevOps / SRE / Platform |
| Quelle | The New Stack ↗ |
| Veröffentlicht | 10.08.2026 UTC |
Sicherheitsmeldung mit Schweregrad noch nicht bewertet. Technische Details im Tab „Originaltext“; empfohlene Schritte in der Checkliste.
Meta released Muse Code on August 5, its first AI coding agent, built on the new Muse Spark 1.2 model. Mark Zuckerberg says it handles “complete software engineering tasks across large repos,” and that big jobs “fan out to separate sub-agents working in parallel.”
Meta’s AI chief Alexandr Wang was blunt about the strategy. The differentiator, he told CNBC, is being “more affordable than either Claude or Codex.”
Standard muse-spark-1.2 runs $1.25 per million input tokens and $4.25 per million output. For comparison, Fable 5, the Anthropic model my Claude Code ran, lists at $10 and $50. Then there’s the “contributor” tier at $0.10 and $0.20, which Wang says is “more than 10x cheaper.”
He’s right, and here’s the fine print. Contributor means “your content may be used for product improvement.” Your code, your prompts, your sessions, all of it becomes Meta training data. And that tier isn’t buried in a settings menu somewhere. It’s the default.
So the question I tested is half of the claim Wang didn’t finish. More affordable, sure. But affordable at what quality? I gave Muse Code and Claude Code the same three coding jobs and tracked every token.
Setting up Muse
Setup was quick. One curl command installs Muse, you log in through your regular Meta account, and it refuses to run until you attach a payment card. (Kimi pulled the same move on me last month. These companies will trust an AI with your codebase but not you with an invoice.) The pricing table sits on the same page where you add the card, which is where I spotted the default tier.
I ran my tests on contributor …