Ollama (Inference Runtime)open-weight
Released: 2024-12-08

Ollama 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 composite
Throughput Speed
42tok/sec
Output streaming throughput
Time To First Token
420ms
Initial chunk server latency
LiveBench Contam-Free
75.0%
September 2026 suite score

Sourced Benchmark Evaluations (2)

Verified performance across authoritative benchmarks with provenance tracking.
BenchmarkStatusScoreTrustDateSource TypeProvenance
ARC-AGI-2Nearing Saturation64.7%
75
2025-01-05independentSource ↗
SWE-bench VerifiedSaturated58.7%
25
2024-12-17independentSource ↗
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.