Meta

Muse Spark Open 816B-A42B

• Released 15 October 2026
• Framework: Superintelligence Scaling Framework
• 4 tests by the provider
• 1 source document

The provider's own tests

• Agentic Misalignment benchmark · The provider · Deception & sabotage

  On our misalignment evaluations, the model showed low propensity for deceptive or unilateral harmful behavior, with a 98.1% honesty rate on the MASK benchmark and a 0% misbehavior rate on the Agentic Misalignment benchmark.

  Meta · 15 October 2026 · Read the source
• PolyGuard (cyber subset), RedCode, and CyberBench 2024 · The provider · Cyber

  The model refused 99.2% of jailbreaks on our internal benchmark and 94.8% when aggregating across PolyGuard (cyber subset), RedCode, and CyberBench 2024.

  Meta · 15 October 2026 · Read the source
• SecureBio's BioTIER refusal benchmark · Run by SecureBio, reported by the provider · Chemical, biological, radiological and nuclear

  It showed strong refusal behavior on both single-turn and multi-turn benchmarks, refusing over 98% of harmful requests on SecureBio's BioTIER refusal benchmark and over 99% on our internal single-turn chemistry refusal benchmark.

  Meta · 15 October 2026 · Read the source
• internal multi-turn agentic benchmarks · The provider · Chemical, biological, radiological and nuclear

  On newly developed internal multi-turn agentic benchmarks, it refused over 97% of harmful biology requests and 98% of harmful chemistry requests.

  Meta · 15 October 2026 · Read the source

No document we read states: whether a capability level was crossed, an independent or government test, a safeguard. No incident on record names this version.

Compare with other versions