Muse Spark 1.1
Meta · 2026-07-09
Muse Spark 1.1 is Meta Superintelligence Labs' July 2026 upgrade to Muse Spark. It is a proprietary multimodal reasoning model built for agentic tasks, with improvements in tool and computer use, coding, multimodal understanding, multi-agent orchestration, and active 1 million token context management. Meta made it available in Thinking mode in the Meta AI app and on meta.ai, and through the public preview Meta Model API with tool and function calling support.
Benchmark scores
| Benchmark | Score |
|---|---|
| AA-Briefcase Elo | 860 |
| AA-Briefcase v1.1 Elo | 851 |
| AA-LCR | 63.3 |
| AA-Omniscience Index | 28.1 |
| AdvancedIF | 79.4 |
| AdvancedIF (rubric mean score) | 94.4 |
| Agent Red Teaming k=100 (CAIS) | 44.8 |
| AHK-Eval | 83.3 |
| AHK-Eval Algorithms | 83.3 |
| AHK-Eval Data Structures | 50.0 |
| AHK-Eval Date and Time | 83.3 |
| AHK-Eval Easy | 91.7 |
| AHK-Eval Hard | 75.0 |
| AHK-Eval Hidden Cases | 84.5 |
| AHK-Eval Mid | 83.3 |
| AHK-Eval Numbers | 83.3 |
| AHK-Eval Parse Success | 97.2 |
| AHK-Eval Regex | 100.0 |
| AHK-Eval Strings | 100.0 |
| AIME 2025 | 100.0 |
| Antidote: Everyday Edition | 1014 |
| APEX (Mercor) | 21.2 |
| APEX-Accounting | 52.6 |
| APEX-Accounting Pass@1 | 7.7 |
| APEX-Agents | 41.9 |
| APEX-Agents 1.1 | 46.2 |
| APEX-Agents 1.1 Pass@1 | 31.8 |
| APEX-SWE | 38.9 |
| APEX-v1-extended | 73.8 |
| Arena Code WebDev — Brand & Marketing | 1536 |