Ollama (Inference Runtime)open-weight
Released: 2024-12-08Ollama Llama 3.3 (Q4_K_M)
Locally hosted 70B parameter model running on single 48GB GPU (or dual RTX 3090/4090s) with minimal quantization loss (<1.2% delta vs FP16).
Intelligence Index
83.8/ 100
Calibrated multi-domain compositeThroughput Speed
42tok/sec
Output streaming throughputTime To First Token
420ms
Initial chunk server latencyLiveBench Contam-Free
75.0%
September 2026 suite scoreSourced Benchmark Evaluations (61)
Verified performance across authoritative benchmarks with provenance tracking.| Benchmark | Status | Score | Trust | Date | Source Type | Provenance |
|---|---|---|---|---|---|---|
| LiveBench (Continuous) | Active | 75% | 94 | 2025-02-01 | independent | Source ↗ |
| ARC-AGI-3 | Active | 54% | 93 | 2025-01-12 | independent | Source ↗ |
| ARC-AGI-2 | Nearing Saturation | 64.7% | 75 | 2025-01-05 | independent | Source ↗ |
| FrontierMath | Active | 23.1% | 98 | 2025-01-02 | independent | Source ↗ |
| PutnamBench | Active | 42.9% | 90 | 2025-01-01 | independent | Source ↗ |
| Omni-MATH | Deprecated | 61.1% | 10 | 2024-12-30 | independent | Source ↗ |
| SWE-Lancer | Deprecated | 52.5% | 20 | 2024-12-30 | independent | Source ↗ |
| Soohak | Active | 62.7% | 98 | 2024-12-28 | independent | Source ↗ |
| FrontierCode | Active | 44.8% | 98 | 2024-12-28 | independent | Source ↗ |
| Humanity's Last Exam | Active | 36.5% | 93 | 2024-12-26 | independent | Source ↗ |
| StructFlowBench | Nearing Saturation | 65.5% | 75 | 2024-12-26 | independent | Source ↗ |
| AIME | Nearing Saturation | 78.4% | 75 | 2024-12-26 | independent | Source ↗ |
| OlympiadBench | Active | 67.6% | 80 | 2024-12-25 | independent | Source ↗ |
| FireBench | Active | 64.8% | 95 | 2024-12-24 | independent | Source ↗ |
| Terminal-Bench | Active | 54% | 80 | 2024-12-24 | independent | Source ↗ |
| MathArena | Active | 66% | 99 | 2024-12-24 | independent | Source ↗ |
| OpenAI-MRCR | Active | 70.4% | 80 | 2024-12-23 | independent | Source ↗ |
| CodeContests | Deprecated | 50.2% | 10 | 2024-12-23 | independent | Source ↗ |
| SimpleQA | Active | 43.6% | 90 | 2024-12-23 | independent | Source ↗ |
| ComplexBench | Nearing Saturation | 63.2% | 80 | 2024-12-22 | independent | Source ↗ |
| SWE-bench Pro | Active | 47.9% | 90 | 2024-12-22 | independent | Source ↗ |
| LongBench-Pro | Active | 68.7% | 80 | 2024-12-22 | independent | Source ↗ |
| MultiPL-E | Deprecated | 71% | 10 | 2024-12-21 | independent | Source ↗ |
| ZeroSCROLLS | Nearing Saturation | 67% | 78 | 2024-12-21 | independent | Source ↗ |
| AgentIF | Nearing Saturation | 52.4% | 80 | 2024-12-20 | independent | Source ↗ |
| MATH | Saturated | 80.9% | 25 | 2024-12-20 | independent | Source ↗ |
| LiveCodeBench | Saturated | 63.3% | 39 | 2024-12-20 | independent | Source ↗ |
| InfiniteBench | Nearing Saturation | 71.2% | 65 | 2024-12-20 | independent | Source ↗ |
| MMLU-Pro | Nearing Saturation | 77.1% | 75 | 2024-12-19 | independent | Source ↗ |
| Mercury | Deprecated | 67.9% | 10 | 2024-12-19 | independent | Source ↗ |
| SCROLLS | Nearing Saturation | 67% | 65 | 2024-12-19 | independent | Source ↗ |
| RULER | Nearing Saturation | 77.1% | 84 | 2024-12-18 | independent | Source ↗ |
| TabMWP | Saturated | 75.9% | 25 | 2024-12-18 | independent | Source ↗ |
| CyberSecEval | Deprecated | 69.5% | 10 | 2024-12-18 | independent | Source ↗ |
| MMLU-Redux | Saturated | 82.1% | 25 | 2024-12-17 | independent | Source ↗ |
| QuALITY | Nearing Saturation | 73.7% | 78 | 2024-12-17 | independent | Source ↗ |
| SWE-bench Verified | Saturated | 60.2% | 25 | 2024-12-17 | independent | Source ↗ |
| MGSM | Nearing Saturation | 78.4% | 65 | 2024-12-16 | independent | Source ↗ |
| LongBench | Nearing Saturation | 72.1% | 65 | 2024-12-16 | independent | Source ↗ |
| SWE-bench | Deprecated | 52.5% | 10 | 2024-12-16 | independent | Source ↗ |
| AGIEval | Saturated | 71.4% | 25 | 2024-12-15 | independent | Source ↗ |
| QASPER | Nearing Saturation | 69.6% | 65 | 2024-12-15 | independent | Source ↗ |
| TruthfulQA | Saturated | 73.7% | 25 | 2024-12-15 | independent | Source ↗ |
| SuperGLUE | Saturated | 84.7% | 25 | 2024-12-14 | independent | Source ↗ |
| GSM-Hard | Deprecated | 79.2% | 10 | 2024-12-14 | independent | Source ↗ |
| MT-Bench | Nearing Saturation | 7.9% | 75 | 2024-12-13 | independent | Source ↗ |
| BIG-Bench Hard | Saturated | 81.3% | 25 | 2024-12-13 | independent | Source ↗ |
| CodeXGLUE | Deprecated | 72.6% | 23 | 2024-12-13 | independent | Source ↗ |
| MBPP+ | Nearing Saturation | 78.7% | 65 | 2024-12-12 | independent | Source ↗ |
| WinoGrande | Saturated | 86.3% | 28 | 2024-12-12 | independent | Source ↗ |
| IFEval | Deprecated | 75.6% | 25 | 2024-12-12 | independent | Source ↗ |
| GPQA | Nearing Saturation | 73.7% | 78 | 2024-12-12 | independent | Source ↗ |
| HumanEval+ | Saturated | 83.4% | 25 | 2024-12-11 | independent | Source ↗ |
| Needle-in-a-Haystack | Saturated | 96.8% | 44 | 2024-12-11 | independent | Source ↗ |
| CommonsenseQA | Saturated | 87.2% | 28 | 2024-12-11 | independent | Source ↗ |
| GSM8K | Saturated | 89.1% | 25 | 2024-12-10 | independent | Source ↗ |
| GLUE | Saturated | 89.6% | 28 | 2024-12-10 | independent | Source ↗ |
| MBPP | Saturated | 81.8% | 25 | 2024-12-10 | independent | Source ↗ |
| HumanEval | Saturated | 88% | 25 | 2024-12-09 | independent | Source ↗ |
| MMLU | Saturated | 85.5% | 25 | 2024-12-09 | independent | Source ↗ |
| ARC-AGI | Saturated | 78% | 25 | 2024-12-09 | independent | Source ↗ |
Local Hardware Execution (Ollama)Hardware Compatible
# 1. Pull and execute model locally with Ollama
ollama run ollama-llama-3-3-70b-q4
# 2. Or invoke via local OpenAI-compatible endpoint
curl http://localhost:11434/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "ollama-llama-3-3-70b-q4",
"messages": [{"role": "user", "content": "Analyze reasoning chains on AIME 2026"}]
}'Evaluation & Sourcing Notes
Scores listed for Ollama Llama 3.3 (Q4_K_M) 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.