Meta AIopen-weight
Released: 2025-04-05Llama 4 Scout
Meta ultra-long-context open-weight workhorse model featuring an unprecedented 10M token context window.
Intelligence Index
89.2/ 100
Calibrated multi-domain compositeThroughput Speed
140tok/sec
Output streaming throughputTime To First Token
310ms
Initial chunk server latencyLiveBench Contam-Free
76.0%
September 2026 suite scoreLiveBench Multi-Domain Evaluation Breakdown
Monthly refreshed contamination-free test setsLogical Reasoning87.0%
Coding & Repo Repair77.5%
Mathematics (AIME)87.0%
Data Analysis & Tables80.0%
Language & Comprehension84.0%
Instruction Following89.0%
Sourced Benchmark Evaluations (61)
Verified performance across authoritative benchmarks with provenance tracking.| Benchmark | Status | Score | Trust | Date | Source Type | Provenance |
|---|---|---|---|---|---|---|
| ARC-AGI-3 | Active | 57.5% | 93 | 2025-05-10 | independent | Source ↗ |
| ARC-AGI-2 | Nearing Saturation | 69% | 75 | 2025-05-03 | independent | Source ↗ |
| FrontierMath | Active | 24.4% | 98 | 2025-04-30 | independent | Source ↗ |
| PutnamBench | Active | 45.2% | 90 | 2025-04-29 | independent | Source ↗ |
| SWE-Lancer | Deprecated | 52.7% | 20 | 2025-04-27 | independent | Source ↗ |
| Omni-MATH | Deprecated | 64.4% | 10 | 2025-04-27 | independent | Source ↗ |
| Soohak | Active | 66.1% | 98 | 2025-04-25 | independent | Source ↗ |
| FrontierCode | Active | 44.9% | 98 | 2025-04-25 | independent | Source ↗ |
| Humanity's Last Exam | Active | 38.9% | 93 | 2025-04-23 | independent | Source ↗ |
| StructFlowBench | Nearing Saturation | 75.6% | 75 | 2025-04-23 | independent | Source ↗ |
| AIME | Nearing Saturation | 82.6% | 75 | 2025-04-23 | independent | Source ↗ |
| OlympiadBench | Active | 71.3% | 80 | 2025-04-22 | independent | Source ↗ |
| MathArena | Active | 69.6% | 99 | 2025-04-21 | independent | Source ↗ |
| FireBench | Active | 74.8% | 95 | 2025-04-21 | independent | Source ↗ |
| Terminal-Bench | Active | 54.3% | 80 | 2025-04-21 | independent | Source ↗ |
| OpenAI-MRCR | Active | 74.9% | 80 | 2025-04-20 | independent | Source ↗ |
| SimpleQA | Active | 46.4% | 90 | 2025-04-20 | independent | Source ↗ |
| CodeContests | Deprecated | 50.4% | 10 | 2025-04-20 | independent | Source ↗ |
| ComplexBench | Nearing Saturation | 73% | 80 | 2025-04-19 | independent | Source ↗ |
| SWE-bench Pro | Active | 48.1% | 90 | 2025-04-19 | independent | Source ↗ |
| LongBench-Pro | Active | 73.1% | 80 | 2025-04-19 | independent | Source ↗ |
| ZeroSCROLLS | Nearing Saturation | 71.4% | 78 | 2025-04-18 | independent | Source ↗ |
| MultiPL-E | Deprecated | 71.3% | 10 | 2025-04-18 | independent | Source ↗ |
| LiveCodeBench | Saturated | 63.6% | 39 | 2025-04-17 | independent | Source ↗ |
| AgentIF | Nearing Saturation | 60.5% | 80 | 2025-04-17 | independent | Source ↗ |
| MATH | Saturated | 85.3% | 25 | 2025-04-17 | independent | Source ↗ |
| InfiniteBench | Nearing Saturation | 75.8% | 65 | 2025-04-17 | independent | Source ↗ |
| Mercury | Deprecated | 68.2% | 10 | 2025-04-16 | independent | Source ↗ |
| SCROLLS | Nearing Saturation | 71.4% | 65 | 2025-04-16 | independent | Source ↗ |
| MMLU-Pro | Nearing Saturation | 82.1% | 75 | 2025-04-16 | independent | Source ↗ |
| RULER | Nearing Saturation | 82.1% | 84 | 2025-04-15 | independent | Source ↗ |
| TabMWP | Saturated | 80% | 25 | 2025-04-15 | independent | Source ↗ |
| CyberSecEval | Deprecated | 69.8% | 10 | 2025-04-15 | independent | Source ↗ |
| SWE-bench Verified | Saturated | 60.5% | 25 | 2025-04-14 | independent | Source ↗ |
| MMLU-Redux | Saturated | 87.4% | 25 | 2025-04-14 | independent | Source ↗ |
| QuALITY | Nearing Saturation | 78.5% | 78 | 2025-04-14 | independent | Source ↗ |
| LongBench | Nearing Saturation | 76.7% | 65 | 2025-04-13 | independent | Source ↗ |
| SWE-bench | Deprecated | 52.7% | 10 | 2025-04-13 | independent | Source ↗ |
| MGSM | Nearing Saturation | 82.6% | 65 | 2025-04-13 | independent | Source ↗ |
| AGIEval | Saturated | 76.1% | 25 | 2025-04-12 | independent | Source ↗ |
| QASPER | Nearing Saturation | 74% | 65 | 2025-04-12 | independent | Source ↗ |
| TruthfulQA | Saturated | 78.5% | 25 | 2025-04-12 | independent | Source ↗ |
| GSM-Hard | Deprecated | 83.5% | 10 | 2025-04-11 | independent | Source ↗ |
| SuperGLUE | Saturated | 90.3% | 25 | 2025-04-11 | independent | Source ↗ |
| CodeXGLUE | Deprecated | 72.9% | 23 | 2025-04-10 | independent | Source ↗ |
| MT-Bench | Nearing Saturation | 9.2% | 75 | 2025-04-10 | independent | Source ↗ |
| BIG-Bench Hard | Saturated | 86.7% | 25 | 2025-04-10 | independent | Source ↗ |
| WinoGrande | Saturated | 92% | 28 | 2025-04-09 | independent | Source ↗ |
| MBPP+ | Nearing Saturation | 79.1% | 65 | 2025-04-09 | independent | Source ↗ |
| IFEval | Deprecated | 87.2% | 25 | 2025-04-09 | independent | Source ↗ |
| GPQA | Nearing Saturation | 78.5% | 78 | 2025-04-09 | independent | Source ↗ |
| Needle-in-a-Haystack | Saturated | 97.8% | 44 | 2025-04-08 | independent | Source ↗ |
| HumanEval+ | Saturated | 83.7% | 25 | 2025-04-08 | independent | Source ↗ |
| CommonsenseQA | Saturated | 92.9% | 28 | 2025-04-08 | independent | Source ↗ |
| MBPP | Saturated | 82.2% | 25 | 2025-04-07 | independent | Source ↗ |
| GSM8K | Saturated | 94% | 25 | 2025-04-07 | independent | Source ↗ |
| GLUE | Saturated | 95.6% | 28 | 2025-04-07 | independent | Source ↗ |
| MMLU | Saturated | 91% | 25 | 2025-04-06 | independent | Source ↗ |
| HumanEval | Saturated | 88.4% | 25 | 2025-04-06 | independent | Source ↗ |
| ARC-AGI | Saturated | 83.2% | 25 | 2025-04-06 | independent | Source ↗ |
| LiveBench (Continuous) | Active | 76% | 94 | 2025-02-01 | independent | Source ↗ |
Local Hardware Execution (Ollama)Hardware Compatible
# 1. Pull and execute model locally with Ollama
ollama run llama-4-scout
# 2. Or invoke via local OpenAI-compatible endpoint
curl http://localhost:11434/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "llama-4-scout",
"messages": [{"role": "user", "content": "Analyze reasoning chains on AIME 2026"}]
}'Evaluation & Sourcing Notes
Scores listed for Llama 4 Scout 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.