Head-to-head
DeepSeek R1 vs DeepSeek V3 (Thinking)
Normalized scores are min-maxed per benchmark across all models we track (0–100). Open the interactive compare view to add benchmarks to the radar chart or pull in more models.
DeepSeek R1
DeepSeekDeepSeek V3 (Thinking)
DeepSeek| Benchmark | DeepSeek R1 | DeepSeek V3 (Thinking) |
|---|---|---|
Chatbot Arena Elo Arena | ||
AIME 2024 AIME | — | |
HumanEval HumanEval | — | |
LiveCodeBench LiveCB | — | |
RULER 128k RULER | — | |
Output Speed Speed | — | |
Time to First Token TTFT | — | |
Rolling Contamination-Controlled Average Rolling Avg | — | |
Rolling Data Analysis Data Analysis | — | |
SimpleQA Verified SimpleQA | — | |
OTIS Mock AIME 2024-2025 OTIS AIME | — | |
ARC-AGI 2 ARC-AGI 2 | — | |
Aider Polyglot Aider | ||
Frontier Composite Frontier | ||
Output Stability Stability | — | |
Format Adherence Format | — | |
Recovery Rate Recovery | — | |
Safety Handling Safety | — |
Methodology matches the main AI Model Analyzer About page.