Meta AIopen-weight
Released: 2026-09-02Muse Spark 1.3
Muse Spark 1.3 is a multimodal reasoning model from Meta for long-running agentic, multi-agent, and coding workflows. It is designed to keep track of information across extended tasks, work through...
Intelligence Index
88.5/ 100
Calibrated multi-domain compositeThroughput Speed
85tok/sec
Output streaming throughputTime To First Token
280ms
Initial chunk server latencyLiveBench Contam-Free
77.9%
September 2026 suite scoreLiveBench Multi-Domain Evaluation Breakdown
Monthly refreshed contamination-free test setsLogical Reasoning81.4%
Coding & Repo Repair79.7%
Mathematics (AIME)78.8%
Data Analysis & Tables75.2%
Language & Comprehension77.0%
Instruction Following77.9%
Sourced Benchmark Evaluations (60)
Verified performance across authoritative benchmarks with provenance tracking.| Benchmark | Status | Score | Trust | Date | Source Type | Provenance |
|---|---|---|---|---|---|---|
| SCROLLS | Nearing Saturation | 70.8% | 65 | 2026-09-12 | independent | Source ↗ |
| StructFlowBench | Nearing Saturation | 66.2% | 75 | 2026-09-12 | independent | Source ↗ |
| Soohak | Active | 62.5% | 98 | 2026-09-12 | independent | Source ↗ |
| SimpleQA | Active | 46% | 90 | 2026-09-12 | independent | Source ↗ |
| InfiniteBench | Nearing Saturation | 75.2% | 65 | 2026-09-12 | independent | Source ↗ |
| LiveCodeBench | Saturated | 69% | 39 | 2026-09-12 | independent | Source ↗ |
| LongBench-Pro | Active | 72.6% | 80 | 2026-09-12 | independent | Source ↗ |
| MATH | Saturated | 80.7% | 25 | 2026-09-12 | independent | Source ↗ |
| MathArena | Active | 65.8% | 99 | 2026-09-12 | independent | Source ↗ |
| Humanity's Last Exam | Active | 37.8% | 93 | 2026-09-12 | independent | Source ↗ |
| RULER | Nearing Saturation | 81.4% | 84 | 2026-09-12 | independent | Source ↗ |
| Mercury | Deprecated | 74% | 10 | 2026-09-12 | independent | Source ↗ |
| MMLU-Pro | Nearing Saturation | 81.4% | 75 | 2026-09-12 | independent | Source ↗ |
| MultiPL-E | Deprecated | 77.4% | 10 | 2026-09-12 | independent | Source ↗ |
| PutnamBench | Active | 42.8% | 90 | 2026-09-12 | independent | Source ↗ |
| OlympiadBench | Active | 67.5% | 80 | 2026-09-12 | independent | Source ↗ |
| Omni-MATH | Deprecated | 60.9% | 10 | 2026-09-12 | independent | Source ↗ |
| OpenAI-MRCR | Active | 74.3% | 80 | 2026-09-12 | independent | Source ↗ |
| Terminal-Bench | Active | 58.9% | 80 | 2026-09-12 | independent | Source ↗ |
| AgentIF | Nearing Saturation | 53% | 80 | 2026-09-12 | independent | Source ↗ |
| AIME | Nearing Saturation | 78.2% | 75 | 2026-09-12 | independent | Source ↗ |
| ZeroSCROLLS | Nearing Saturation | 70.8% | 78 | 2026-09-12 | independent | Source ↗ |
| ARC-AGI-2 | Nearing Saturation | 66.9% | 75 | 2026-09-12 | independent | Source ↗ |
| ARC-AGI-3 | Active | 55.8% | 93 | 2026-09-12 | independent | Source ↗ |
| CodeContests | Deprecated | 54.7% | 10 | 2026-09-12 | independent | Source ↗ |
| TabMWP | Saturated | 75.7% | 25 | 2026-09-12 | independent | Source ↗ |
| ComplexBench | Nearing Saturation | 63.9% | 80 | 2026-09-12 | independent | Source ↗ |
| CyberSecEval | Deprecated | 75.7% | 10 | 2026-09-12 | independent | Source ↗ |
| FireBench | Active | 65.4% | 95 | 2026-09-12 | independent | Source ↗ |
| FrontierCode | Active | 48.8% | 98 | 2026-09-12 | independent | Source ↗ |
| FrontierMath | Active | 23% | 98 | 2026-09-12 | independent | Source ↗ |
| SWE-Lancer | Deprecated | 57.2% | 20 | 2026-09-12 | independent | Source ↗ |
| SWE-bench Pro | Active | 52.1% | 90 | 2026-09-12 | independent | Source ↗ |
| QuALITY | Nearing Saturation | 77.9% | 78 | 2026-09-11 | independent | Source ↗ |
| MMLU-Redux | Saturated | 86.7% | 25 | 2026-09-11 | independent | Source ↗ |
| SWE-bench Verified | Saturated | 65.6% | 25 | 2026-09-11 | independent | Source ↗ |
| LongBench | Nearing Saturation | 76.1% | 65 | 2026-09-10 | independent | Source ↗ |
| SWE-bench | Deprecated | 57.2% | 10 | 2026-09-10 | independent | Source ↗ |
| MGSM | Nearing Saturation | 78.2% | 65 | 2026-09-10 | independent | Source ↗ |
| TruthfulQA | Saturated | 77.9% | 25 | 2026-09-09 | independent | Source ↗ |
| QASPER | Nearing Saturation | 73.5% | 65 | 2026-09-09 | independent | Source ↗ |
| AGIEval | Saturated | 73.8% | 25 | 2026-09-09 | independent | Source ↗ |
| SuperGLUE | Saturated | 87.5% | 25 | 2026-09-08 | independent | Source ↗ |
| GSM-Hard | Deprecated | 79% | 10 | 2026-09-08 | independent | Source ↗ |
| MT-Bench | Nearing Saturation | 8% | 75 | 2026-09-07 | independent | Source ↗ |
| BIG-Bench Hard | Saturated | 84.1% | 25 | 2026-09-07 | independent | Source ↗ |
| CodeXGLUE | Deprecated | 79.1% | 23 | 2026-09-07 | independent | Source ↗ |
| WinoGrande | Saturated | 89.2% | 28 | 2026-09-06 | independent | Source ↗ |
| MBPP+ | Nearing Saturation | 85.8% | 65 | 2026-09-06 | independent | Source ↗ |
| IFEval | Deprecated | 76.3% | 25 | 2026-09-06 | independent | Source ↗ |
| GPQA | Nearing Saturation | 77.9% | 78 | 2026-09-06 | independent | Source ↗ |
| HumanEval+ | Saturated | 90.8% | 25 | 2026-09-05 | independent | Source ↗ |
| Needle-in-a-Haystack | Saturated | 97.7% | 44 | 2026-09-05 | independent | Source ↗ |
| CommonsenseQA | Saturated | 90.1% | 28 | 2026-09-05 | independent | Source ↗ |
| GSM8K | Saturated | 88.9% | 25 | 2026-09-04 | independent | Source ↗ |
| GLUE | Saturated | 92.7% | 28 | 2026-09-04 | independent | Source ↗ |
| MBPP | Saturated | 89.1% | 25 | 2026-09-04 | independent | Source ↗ |
| HumanEval | Saturated | 95.9% | 25 | 2026-09-03 | independent | Source ↗ |
| MMLU | Saturated | 90.3% | 25 | 2026-09-03 | independent | Source ↗ |
| ARC-AGI | Saturated | 80.7% | 25 | 2026-09-03 | independent | Source ↗ |
Local Hardware Execution (Ollama)Hardware Compatible
# 1. Pull and execute model locally with Ollama
ollama run meta-muse-spark-1-3
# 2. Or invoke via local OpenAI-compatible endpoint
curl http://localhost:11434/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "meta-muse-spark-1-3",
"messages": [{"role": "user", "content": "Analyze reasoning chains on AIME 2026"}]
}'Evaluation & Sourcing Notes
Scores listed for Muse Spark 1.3 represent verified evaluations extracted from official research papers, independent evaluation suites (HELM, LMSYS, OpenCompass, LiveBench), and verified audit reports.
All benchmarks marked as saturated or deprecated reflect historical performance where the benchmark no longer provides active discriminative power.