Compare · ModelsLive · 2 picked · head to head
Kimi K2.7 Code vs Kimi K3
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Kimi K3 wins on 19/19 benchmarks
Kimi K3 wins 19 of 19 shared benchmarks. Leads in speed · agentic · knowledge.
Category leads
speed·Kimi K3agentic·Kimi K3knowledge·Kimi K3coding·Kimi K3math·Kimi K3reasoning·Kimi K3general·Kimi K3
Hype vs Reality
Attention vs performance
Kimi K2.7 Code
#77 by perf·#18 by attention
Kimi K3
#54 by perf·#18 by attention
Best value
Kimi K2.7 Code
3.6x better value than Kimi K3
Kimi K2.7 Code
27.8 pts/$
$2.01/M
Kimi K3
7.7 pts/$
$7.72/M
Vendor risk
Who is behind the model
Moonshot AI
$50.0B·Tier 1
Moonshot AI
$50.0B·Tier 1
Head to head
19 benchmarks · 2 models
Kimi K2.7 CodeKimi K3
Artificial Analysis · CritPt
Kimi K3 leads by +13.4
Kimi K2.7 Code
10.0
Kimi K3
23.4
Artificial Analysis · GDPval
Kimi K3 leads by +24.8
Kimi K2.7 Code
26.9
Kimi K3
51.7
Artificial Analysis · GPQA Diamond
Kimi K3 leads by +3.9
Kimi K2.7 Code
89.6
Kimi K3
93.5
Artificial Analysis · Humanity's Last Exam
Kimi K3 leads by +11.9
Kimi K2.7 Code
35.0
Kimi K3
46.9
Artificial Analysis · Long Context Reasoning
Kimi K3 leads by +9.4
Kimi K2.7 Code
79.3
Kimi K3
88.7
Artificial Analysis · Quality Index
Kimi K3 leads by +17.8
Kimi K2.7 Code
25.8
Kimi K3
43.6
Artificial Analysis · SciCode
Kimi K3 leads by +11.7
Kimi K2.7 Code
47.8
Kimi K3
59.5
APEX-Agents
Kimi K3 leads by +13.0
APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments.
Kimi K2.7 Code
37.6
Kimi K3
50.6
Chess Puzzles
Kimi K3 leads by +24.2
Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities.
Kimi K2.7 Code
9.5
Kimi K3
33.7
Deepswe
Kimi K3 leads by +38.0
Kimi K2.7 Code
30.5
Kimi K3
68.5
Frontiercode
Kimi K3 leads by +14.1
Kimi K2.7 Code
30.1
Kimi K3
44.2
FrontierMath-Tier-4-v2-Private
Kimi K3 leads by +26.8
Kimi K2.7 Code
12.2
Kimi K3
39.0
FrontierMath-Tiers-1-3-v2-Private
Kimi K3 leads by +18.1
Kimi K2.7 Code
54.0
Kimi K3
72.2
GPQA diamond
Kimi K3 leads by +7.0
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
Kimi K2.7 Code
83.8
Kimi K3
90.8
OTIS Mock AIME 2024-2025
Kimi K3 leads by +1.7
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Kimi K2.7 Code
95.5
Kimi K3
97.2
SimpleBench
Kimi K3 leads by +3.4
SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking.
Kimi K2.7 Code
49.5
Kimi K3
52.8
SimpleQA Verified
Kimi K3 leads by +14.1
SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information.
Kimi K2.7 Code
36.5
Kimi K3
50.6
Surface Evolver Bench
Kimi K3 leads by +46.3
Kimi K2.7 Code
48.8
Kimi K3
95.0
WeirdML
Kimi K3 leads by +28.4
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Kimi K2.7 Code
54.1
Kimi K3
82.6
Full benchmark table
| Benchmark | Kimi K2.7 Code | Kimi K3 |
|---|---|---|
Artificial Analysis · CritPt | 10.0 | 23.4 |
Artificial Analysis · GDPval | 26.9 | 51.7 |
Artificial Analysis · GPQA Diamond | 89.6 | 93.5 |
Artificial Analysis · Humanity's Last Exam | 35.0 | 46.9 |
Artificial Analysis · Long Context Reasoning | 79.3 | 88.7 |
Artificial Analysis · Quality Index | 25.8 | 43.6 |
Artificial Analysis · SciCode | 47.8 | 59.5 |
APEX-Agents APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments. | 37.6 | 50.6 |
Chess Puzzles Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities. | 9.5 | 33.7 |
Deepswe | 30.5 | 68.5 |
Frontiercode | 30.1 | 44.2 |
FrontierMath-Tier-4-v2-Private | 12.2 | 39.0 |
FrontierMath-Tiers-1-3-v2-Private | 54.0 | 72.2 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 83.8 | 90.8 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 95.5 | 97.2 |
SimpleBench SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking. | 49.5 | 52.8 |
SimpleQA Verified SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information. | 36.5 | 50.6 |
Surface Evolver Bench | 48.8 | 95.0 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 54.1 | 82.6 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $0.67 | $3.35 | 262K tokens (~131 books) | $13.41 | |
| $0.45 | $15.00 | 1.0M tokens (~524 books) | $40.88 |