EngineersofAI LogoCoursesBlogResearch LabAI LettersThe LabCode BankInteractive 3DKodrEarnestJobs
Skip to main content
Interactive 3D/LLM Benchmark Explorer
Benchmark Radar
GPT-4o
Claude-3.5
Llama-3 70B
MMLUHumanEvalMATHHellaSwagTruthfulQAHELM255075100
Score Breakdown
MMLUknowledge
GPT-4o
86.4
Claude-3.5
88.7
Llama-3 70B
82.0
HumanEvalcoding
GPT-4o
90.2
Claude-3.5
92.0
Llama-3 70B
81.7
MATHreasoning
GPT-4o
72.6
Claude-3.5
71.1
Llama-3 70B
50.4
HellaSwagcommonsense
GPT-4o
95.3
Claude-3.5
95.8
Llama-3 70B
93.2
TruthfulQAfactuality
GPT-4o
59.0
Claude-3.5
63.2
Llama-3 70B
52.3
HELMoverall
GPT-4o
81.2
Claude-3.5
82.7
Llama-3 70B
73.1
Overall Ranking (selected benchmarks)
#1Claude-3.5
82.2
#2GPT-4o
80.8
#3Gemini 1.5
74.1
#4Llama-3 70B
72.1
#5Mistral Large
68.4
Controls
Models
Category Filter
No single model dominates all benchmarks. Claude-3.5 leads on knowledge and coding; GPT-4o on commonsense. Hover benchmark rows for descriptions. Select category filter to isolate one capability area.

LLM Benchmark Explorer - Interactive Visualization

No single LLM dominates every benchmark. MMLU tests broad factual knowledge across 57 subjects. HumanEval measures Python coding ability on 164 problems. MATH tests multi-step mathematical reasoning. HellaSwag evaluates commonsense scenario completion. TruthfulQA probes whether models avoid common misconceptions. HELM provides a holistic aggregate across 42 diverse scenarios.

  • Claude-3.5 leads on MMLU knowledge (88.7) and TruthfulQA factuality (63.2) as of 2024
  • GPT-4o and Claude-3.5 are near-equivalent on most benchmarks; differences matter by task category
  • Llama-3 70B closes the gap on coding and commonsense while remaining fully open-source
  • Benchmark rankings do not always predict real-world task performance for specific use cases

Part of the EngineersOfAI Interactive 3D - free interactive visualizations covering every major concept in machine learning and AI engineering. Hover any element for a plain-English explanation. No code required.