Compare · ModelsLive · 2 picked · head to head
Kimi K2 Thinking vs GPT-5 Mini
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Kimi K2 Thinking wins on 9/16 benchmarks
Kimi K2 Thinking wins 9 of 16 shared benchmarks. Leads in knowledge · coding · reasoning.
Category leads
math·GPT-5 Miniknowledge·Kimi K2 Thinkingcoding·Kimi K2 Thinkingreasoning·Kimi K2 Thinkinglanguage·GPT-5 Mini
Hype vs Reality
Attention vs performance
Kimi K2 Thinking
#79 by perf·no signal
GPT-5 Mini
#65 by perf·no signal
Best value
GPT-5 Mini
1.4x better value than Kimi K2 Thinking
Kimi K2 Thinking
34.4 pts/$
$1.55/M
GPT-5 Mini
49.8 pts/$
$1.13/M
Vendor risk
Who is behind the model
moonshotai
private · undisclosed
OpenAI
$840.0B·Tier 1
Head to head
16 benchmarks · 2 models
Kimi K2 ThinkingGPT-5 Mini
FrontierMath-2025-02-28-Private
GPT-5 Mini leads by +5.8
FrontierMath (Feb 2025) · original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning.
Kimi K2 Thinking
21.4
GPT-5 Mini
27.2
FrontierMath-Tier-4-2025-07-01-Private
GPT-5 Mini leads by +6.2
FrontierMath Tier 4 (Jul 2025) · the most challenging tier of frontier mathematics, containing problems that push the absolute limits of AI mathematical reasoning.
Kimi K2 Thinking
0.1
GPT-5 Mini
6.3
GPQA diamond
Kimi K2 Thinking leads by +12.3
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
Kimi K2 Thinking
79.0
GPT-5 Mini
66.7
LiveBench · Agentic Coding
Kimi K2 Thinking leads by +3.3
Kimi K2 Thinking
38.3
GPT-5 Mini
35.0
LiveBench · Coding
GPT-5 Mini leads by +8.6
Kimi K2 Thinking
67.4
GPT-5 Mini
76.1
LiveBench · Data Analysis
Kimi K2 Thinking leads by +2.7
Kimi K2 Thinking
52.3
GPT-5 Mini
49.6
LiveBench · If
GPT-5 Mini leads by +2.2
Kimi K2 Thinking
62.0
GPT-5 Mini
64.2
LiveBench · Language
GPT-5 Mini leads by +2.7
Kimi K2 Thinking
66.5
GPT-5 Mini
69.2
LiveBench · Mathematics
Kimi K2 Thinking leads by +6.7
Kimi K2 Thinking
81.1
GPT-5 Mini
74.4
LiveBench · Overall
Kimi K2 Thinking leads by +0.6
Kimi K2 Thinking
61.6
GPT-5 Mini
61.0
LiveBench · Reasoning
Kimi K2 Thinking leads by +4.8
Kimi K2 Thinking
63.5
GPT-5 Mini
58.6
OTIS Mock AIME 2024-2025
GPT-5 Mini leads by +3.6
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Kimi K2 Thinking
83.0
GPT-5 Mini
86.7
SimpleQA Verified
Kimi K2 Thinking leads by +10.6
SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information.
Kimi K2 Thinking
31.6
GPT-5 Mini
21.0
SWE-Bench Verified (Bash Only)
Kimi K2 Thinking leads by +3.6
SWE-Bench Verified (Bash Only) · a curated subset of SWE-bench where models fix real Python repository bugs using only bash commands, no agent frameworks.
Kimi K2 Thinking
63.4
GPT-5 Mini
59.8
Terminal Bench
Kimi K2 Thinking leads by +0.9
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.
Kimi K2 Thinking
35.7
GPT-5 Mini
34.8
WeirdML
GPT-5 Mini leads by +9.9
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Kimi K2 Thinking
42.8
GPT-5 Mini
52.7
Full benchmark table
| Benchmark | Kimi K2 Thinking | GPT-5 Mini |
|---|---|---|
FrontierMath-2025-02-28-Private FrontierMath (Feb 2025) · original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning. | 21.4 | 27.2 |
FrontierMath-Tier-4-2025-07-01-Private FrontierMath Tier 4 (Jul 2025) · the most challenging tier of frontier mathematics, containing problems that push the absolute limits of AI mathematical reasoning. | 0.1 | 6.3 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 79.0 | 66.7 |
LiveBench · Agentic Coding | 38.3 | 35.0 |
LiveBench · Coding | 67.4 | 76.1 |
LiveBench · Data Analysis | 52.3 | 49.6 |
LiveBench · If | 62.0 | 64.2 |
LiveBench · Language | 66.5 | 69.2 |
LiveBench · Mathematics | 81.1 | 74.4 |
LiveBench · Overall | 61.6 | 61.0 |
LiveBench · Reasoning | 63.5 | 58.6 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 83.0 | 86.7 |
SimpleQA Verified SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information. | 31.6 | 21.0 |
SWE-Bench Verified (Bash Only) SWE-Bench Verified (Bash Only) · a curated subset of SWE-bench where models fix real Python repository bugs using only bash commands, no agent frameworks. | 63.4 | 59.8 |
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. | 35.7 | 34.8 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 42.8 | 52.7 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $0.60 | $2.50 | 262K tokens (~131 books) | $10.75 | |
| $0.25 | $2.00 | 400K tokens (~200 books) | $6.88 |