Compare · ModelsLive · 2 picked · head to head
Grok 4.20 vs Kimi K2.5
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Grok 4.20 wins on 8/10 benchmarks
Grok 4.20 wins 8 of 10 shared benchmarks. Leads in reasoning · knowledge · general.
Category leads
reasoning·Grok 4.20knowledge·Grok 4.20general·Grok 4.20math·Grok 4.20coding·Grok 4.20
Hype vs Reality
Attention vs performance
Grok 4.20
#136 by perf·#16 by attention
Kimi K2.5
#129 by perf·#18 by attention
Best value
Kimi K2.5
1.4x better value than Grok 4.20
Grok 4.20
26.2 pts/$
$1.88/M
Kimi K2.5
37.0 pts/$
$1.35/M
Vendor risk
Who is behind the model
xAI
$250.0B·Tier 1
Moonshot AI
$50.0B·Tier 1
Head to head
10 benchmarks · 2 models
Grok 4.20Kimi K2.5
ARC-AGI
Grok 4.20 leads by +24.2
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
Grok 4.20
89.5
Kimi K2.5
65.3
ARC-AGI-2
Grok 4.20 leads by +53.3
ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data.
Grok 4.20
65.1
Kimi K2.5
11.8
Chess Puzzles
Grok 4.20 leads by +17.9
Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities.
Grok 4.20
25.3
Kimi K2.5
7.4
Cl Bench
Grok 4.20 leads by +2.9
Grok 4.20
22.2
Kimi K2.5
19.3
Cl Bench Life
Kimi K2.5 leads by +1.3
Grok 4.20
11.9
Kimi K2.5
13.2
GPQA diamond
Grok 4.20 leads by +2.3
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
Grok 4.20
85.8
Kimi K2.5
83.5
OTIS Mock AIME 2024-2025
Grok 4.20 leads by +0.0
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Grok 4.20
92.2
Kimi K2.5
92.2
SimpleQA Verified
Kimi K2.5 leads by +4.1
SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information.
Grok 4.20
30.2
Kimi K2.5
34.3
Terminal Bench
Grok 4.20 leads by +14.1
Terminal-Bench 2.0 · evaluates AI agents on real terminal-based coding tasks · writing scripts, debugging, running tests, and managing projects entirely through command-line interaction. Tests both code quality and terminal fluency. Claude Opus 4.7 scores 69.4%, demonstrating significant agentic terminal competence.
Grok 4.20
57.3
Kimi K2.5
43.2
WeirdML
Grok 4.20 leads by +6.7
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Grok 4.20
52.3
Kimi K2.5
45.6
Full benchmark table
| Benchmark | Grok 4.20 | Kimi K2.5 |
|---|---|---|
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | 89.5 | 65.3 |
ARC-AGI-2 ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data. | 65.1 | 11.8 |
Chess Puzzles Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities. | 25.3 | 7.4 |
Cl Bench | 22.2 | 19.3 |
Cl Bench Life | 11.9 | 13.2 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 85.8 | 83.5 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 92.2 | 92.2 |
SimpleQA Verified SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information. | 30.2 | 34.3 |
Terminal Bench Terminal-Bench 2.0 · evaluates AI agents on real terminal-based coding tasks · writing scripts, debugging, running tests, and managing projects entirely through command-line interaction. Tests both code quality and terminal fluency. Claude Opus 4.7 scores 69.4%, demonstrating significant agentic terminal competence. | 57.3 | 43.2 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 52.3 | 45.6 |
Pricing · per 1M tokens · projected $/mo at 10M tokens