Compare · ModelsLive · 2 picked · head to head
GLM 5 vs Kimi K2.5
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Kimi K2.5 wins on 12/20 benchmarks
Kimi K2.5 wins 12 of 20 shared benchmarks. Leads in reasoning · knowledge · general.
Category leads
agentic·GLM 5reasoning·Kimi K2.5knowledge·Kimi K2.5general·Kimi K2.5math·Kimi K2.5language·Kimi K2.5coding·GLM 5
Hype vs Reality
Attention vs performance
GLM 5
#79 by perf·#3 by attention
Kimi K2.5
#122 by perf·#17 by attention
Best value
GLM 5
1.2x better value than Kimi K2.5
GLM 5
43.7 pts/$
$1.26/M
Kimi K2.5
37.0 pts/$
$1.35/M
Vendor risk
Who is behind the model
z-ai
private · undisclosed
Moonshot AI
$18.0B·Tier 1
Head to head
20 benchmarks · 2 models
GLM 5Kimi K2.5
APEX-Agents
GLM 5 leads by +2.8
APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments.
GLM 5
17.2
Kimi K2.5
14.4
ARC-AGI
Kimi K2.5 leads by +20.7
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
GLM 5
44.7
Kimi K2.5
65.3
ARC-AGI-2
Kimi K2.5 leads by +7.0
ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data.
GLM 5
4.9
Kimi K2.5
11.8
Chess Puzzles
Kimi K2.5 leads by +2.1
Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities.
GLM 5
5.3
Kimi K2.5
7.4
Cl Bench
Kimi K2.5 leads by +0.6
GLM 5
18.7
Kimi K2.5
19.3
FrontierMath-2025-02-28-Private
Kimi K2.5 leads by +20.1
FrontierMath (Feb 2025) · original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning.
GLM 5
28.8
Kimi K2.5
49.0
FrontierMath-Tier-4-2025-07-01-Private
Kimi K2.5 leads by +3.5
FrontierMath Tier 4 (Jul 2025) · the most challenging tier of frontier mathematics, containing problems that push the absolute limits of AI mathematical reasoning.
GLM 5
3.5
Kimi K2.5
7.0
GPQA diamond
GLM 5 leads by +0.3
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
GLM 5
83.8
Kimi K2.5
83.5
OpenCompass · AIME2025
GLM 5 leads by +3.9
GLM 5
95.8
Kimi K2.5
91.9
OpenCompass · GPQA-Diamond
Kimi K2.5 leads by +2.8
GLM 5
85.3
Kimi K2.5
88.1
OpenCompass · HLE
Kimi K2.5 leads by +0.5
GLM 5
28.1
Kimi K2.5
28.6
OpenCompass · IFEval
Kimi K2.5 leads by +0.7
GLM 5
93.2
Kimi K2.5
93.9
OpenCompass · LiveCodeBenchV6
GLM 5 leads by +5.6
GLM 5
86.2
Kimi K2.5
80.6
OpenCompass · MMLU-Pro
Kimi K2.5 leads by +1.0
GLM 5
85.2
Kimi K2.5
86.2
OTIS Mock AIME 2024-2025
Kimi K2.5 leads by +12.2
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
GLM 5
80.0
Kimi K2.5
92.2
PostTrainBench
GLM 5 leads by +3.6
GLM 5
13.9
Kimi K2.5
10.3
SimpleBench
GLM 5 leads by +7.7
SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking.
GLM 5
43.8
Kimi K2.5
36.2
SWE-Bench verified
Kimi K2.5 leads by +1.7
SWE-bench Verified · 500 human-validated tasks from 12 real Python repositories (Django, Flask, scikit-learn, sympy, and others). Each task requires the model to produce a git patch that resolves a real GitHub issue and passes the test suite. The verified subset eliminates ambiguous tasks from the original SWE-bench. Claude Mythos Preview leads at 93.9%, crossing 90% for the first time in 2026. Opus 4.6 scores 80.8%. The benchmark remains the most-cited evaluation for code-generation capability.
GLM 5
72.1
Kimi K2.5
73.8
Terminal Bench
GLM 5 leads by +9.2
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.
GLM 5
52.4
Kimi K2.5
43.2
WeirdML
GLM 5 leads by +2.6
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
GLM 5
48.2
Kimi K2.5
45.6
Full benchmark table
| Benchmark | GLM 5 | Kimi K2.5 |
|---|---|---|
APEX-Agents APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments. | 17.2 | 14.4 |
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | 44.7 | 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. | 4.9 | 11.8 |
Chess Puzzles Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities. | 5.3 | 7.4 |
Cl Bench | 18.7 | 19.3 |
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. | 28.8 | 49.0 |
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. | 3.5 | 7.0 |
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 | 83.5 |
OpenCompass · AIME2025 | 95.8 | 91.9 |
OpenCompass · GPQA-Diamond | 85.3 | 88.1 |
OpenCompass · HLE | 28.1 | 28.6 |
OpenCompass · IFEval | 93.2 | 93.9 |
OpenCompass · LiveCodeBenchV6 | 86.2 | 80.6 |
OpenCompass · MMLU-Pro | 85.2 | 86.2 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 80.0 | 92.2 |
PostTrainBench | 13.9 | 10.3 |
SimpleBench SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking. | 43.8 | 36.2 |
SWE-Bench verified SWE-bench Verified · 500 human-validated tasks from 12 real Python repositories (Django, Flask, scikit-learn, sympy, and others). Each task requires the model to produce a git patch that resolves a real GitHub issue and passes the test suite. The verified subset eliminates ambiguous tasks from the original SWE-bench. Claude Mythos Preview leads at 93.9%, crossing 90% for the first time in 2026. Opus 4.6 scores 80.8%. The benchmark remains the most-cited evaluation for code-generation capability. | 72.1 | 73.8 |
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. | 52.4 | 43.2 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 48.2 | 45.6 |
Pricing · per 1M tokens · projected $/mo at 10M tokens