Reasoning
Like-for-like- GLM-5
- 60.8
- Kimi K2.5
- 61.0
- Weighted basis
- 1 vs 1 rows
- Reading
- Kimi K2.5 leads
GLM-5 vs Kimi K2.5
Updated September 3, 2026. Public scores include evidence status and uncertainty. They are not guarantees for a specific workload.
Decision reading
GLM-5 has the higher public score estimate, 65.68 versus 58.81, but the 90% score intervals overlap. Treat that as a lead, not a settled winner.
34 results are shared. Category rows based on different benchmark sets are marked directional and do not name a winner.
Which one for your work
Recommendations appear only when a shared evidence basis or an explicit operating constraint supports the call. Secondary and unsupported use cases stay disclosed below the initial list.
Long documents
Prompts that approach the documented context limit
Kimi K2.5
Kimi K2.5 has the larger documented context window.
Confidence: documented
Chat turn cost
1K fresh input + 500 output tokens
Kimi K2.5
Kimi K2.5 has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates
Show secondary and unsupported calls
Repository review cost
50K fresh input + 3K output tokens
Kimi K2.5
Kimi K2.5 has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates
Coding work
Code generation, repair, and software-engineering tasks
Not enough matched evidence
The category averages use different weighted benchmark sets, so they are directional rather than like-for-like.
Confidence: limited
Agentic work
Tool use, computer use, and multi-step task completion
Not enough matched evidence
The category averages use different weighted benchmark sets, so they are directional rather than like-for-like.
Confidence: limited
Cache-heavy agent loop cost
200K cached + 20K fresh input + 10K output tokens
Not enough matched evidence
The page does not recommend a cost winner because at least one model cannot fit the stated workload in one request. GLM-5 does not fit this workload in one request. GLM-5 has no published cached-input rate, so cached tokens use its listed input rate. Kimi K2.5 has no published cached-input rate, so cached tokens use its listed input rate.
Confidence: rate-fallback
What is actually comparable
Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.
2 GLM-5 only
34 Shared
11 Kimi K2.5 only
- Shared results
- 34
- GLM-5 only
- 2
- Kimi K2.5 only
- 11
- Like-for-like categories
- 5 / 8
2 categories use different evidence sets. Those rows remain visible for coverage context but do not name a winner.
Category results, on a stated basis
Each row states whether both averages use the same weighted benchmark set. Directional and not-comparable rows remain visible, but they never receive a winner in this template.
Knowledge
Like-for-like- GLM-5
- 66.4
- Kimi K2.5
- 56.9
- Weighted basis
- 4 vs 4 rows
- Reading
- GLM-5 leads
Math
Like-for-like- GLM-5
- 56.3
- Kimi K2.5
- 60.6
- Weighted basis
- 4 vs 4 rows
- Reading
- Kimi K2.5 leads
Multilingual
Like-for-like- GLM-5
- 83.1
- Kimi K2.5
- 82.3
- Weighted basis
- 1 vs 1 rows
- Reading
- GLM-5 leads
Instruction following
Like-for-like- GLM-5
- 92.6
- Kimi K2.5
- 93.9
- Weighted basis
- 1 vs 1 rows
- Reading
- Kimi K2.5 leads
Agentic
Directional only- GLM-5
- 56.2
- Kimi K2.5
- 55.0
- Weighted basis
- 1 vs 2 rows
- Reading
- Directional only
Coding
Directional only- GLM-5
- 66.3
- Kimi K2.5
- 59.4
- Weighted basis
- 3 vs 4 rows
- Reading
- Directional only
Multimodal
Not comparable- GLM-5
- Not measured
- Kimi K2.5
- 78.5
- Weighted basis
- 0 vs 1 rows
- Reading
- Not comparable
| Category | GLM-5 | Kimi K2.5 | Weighted basis | Reading |
|---|---|---|---|---|
| Reasoning | 60.8 | 61.0 | Like-for-like1 vs 1 rows | Kimi K2.5 leads |
| Knowledge | 66.4 | 56.9 | Like-for-like4 vs 4 rows | GLM-5 leads |
| Math | 56.3 | 60.6 | Like-for-like4 vs 4 rows | Kimi K2.5 leads |
| Multilingual | 83.1 | 82.3 | Like-for-like1 vs 1 rows | GLM-5 leads |
| Instruction following | 92.6 | 93.9 | Like-for-like1 vs 1 rows | Kimi K2.5 leads |
| Agentic | 56.2 | 55.0 | Directional only1 vs 2 rows | Directional only |
| Coding | 66.3 | 59.4 | Directional only3 vs 4 rows | Directional only |
| Multimodal | Not measured | 78.5 | Not comparable0 vs 1 rows | Not comparable |
Shape of the matched evidence
Only shared public evidence is shown. Sparse evidence stays a ruled list rather than being closed into a radar shape.
Too few matched category axes support a radar. The ruled list below shows only shared benchmark results; positions use each benchmark’s normalized display scale when available.
HLE
Knowledge
FrontierMath v2 (Tiers 1-3)
Math
Terminal-Bench 2.0
Agentic
SWE-bench Pro
Coding
SWE-Rebench
Coding
What each workload costs
Three fixed token mixes turn per-token rates into comparable decisions. Each scenario states context fit and whether cached input had to fall back to the published list-input rate.
Chat turn
1K fresh input + 500 output tokens
- GLM-5
- $0.0026
- Fits in one request
- Kimi K2.5
- $0.0021
- Fits in one request
Kimi K2.5 has the lower modeled cost
Costs use the listed standard API rates.
Repository review
50K fresh input + 3K output tokens
- GLM-5
- $0.0596
- Fits in one request
- Kimi K2.5
- $0.039
- Fits in one request
Kimi K2.5 has the lower modeled cost
Costs use the listed standard API rates.
Cache-heavy agent loop
200K cached + 20K fresh input + 10K output tokens
- GLM-5
- $0.252
- Does not fit in one request
- Cached input priced at the published list-input rate
- Kimi K2.5
- $0.162
- Fits in one request
- Cached input priced at the published list-input rate
GLM-5 does not fit this workload in one request. GLM-5 has no published cached-input rate, so cached tokens use its listed input rate. Kimi K2.5 has no published cached-input rate, so cached tokens use its listed input rate.
Specification differences
Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.
SpecificationGLM-5Kimi K2.5
Context window
Maximum documented context; output-token limits may be lower.
GLM-5
200K
Kimi K2.5
256K
API model ID
GLM-5
Not sourced
Kimi K2.5
Not sourced
Cached-input rate
A missing cached-input rate falls back to the listed input rate only in the stated workload estimate.
GLM-5
Not published
Kimi K2.5
Not published
Documented inputs
GLM-5
Not sourced
Kimi K2.5
Not sourced
Documented outputs
GLM-5
Not sourced
Kimi K2.5
Not sourced
Provider availability
GLM-5
Not sourced
Kimi K2.5
Not sourced
Reasoning profile
GLM-5
Non-Reasoning
Kimi K2.5
Non-Reasoning
Weight access
GLM-5
Open Weight
Kimi K2.5
Open Weight
License
GLM-5
Open Weight
Kimi K2.5
Open Weight
Release date
GLM-5
2026-03-01
Kimi K2.5
2026-02-01
If you already use one of these models
- Deployment change
- The models list different providers, so authentication, endpoint behavior, limits, and feature support may change.
- Quality signal
- GLM-5 has the higher public score estimate, 65.68 versus 58.81, but the 90% score intervals overlap.
- Workload cost
- Repository review: $0.0596 vs $0.039. Cache-heavy agent loop: $0.252 vs $0.162.
- Context tradeoff
- Kimi K2.5 has the larger documented window (256K).
Run the same representative tasks against both endpoints before changing production traffic.
Self-host vs API cost
Estimates at 50,000 req/day · 1000 tokens/req average.
GLM-5
API / mo$3,150
Self-host / moNot listed
Break-even—
Proprietary model — self-hosting not applicable.
Kimi K2.5
API / mo$2,700
Self-host / mo$5,221
Break-even132M/day
Benchmark evidence
The full public result ledger is available for audit without forcing a wide desktop table onto a phone.
Browse raw public benchmark evidence47 rows
Agentic
Terminal-Bench 2.0
GLM-556.2%SourceKimi K2.550.8%SourceGLM-5 leads this result
Claw-Eval
GLM-557.7%SourceKimi K2.552.3%SourceGLM-5 leads this result
QwenClawBench
Shared sourceGLM-554.1%Kimi K2.554.3%Kimi K2.5 leads this result
τ³-bench results
Shared sourceGLM-565.6%Kimi K2.565.7%Kimi K2.5 leads this result
DeepPlanning
Shared sourceGLM-514.6%Kimi K2.514.4%GLM-5 leads this result
Toolathlon
GLM-538%SourceKimi K2.527.8%SourceGLM-5 leads this result
MCP Atlas
GLM-531.1%SourceKimi K2.529.5%SourceGLM-5 leads this result
MCP-Tasks
Shared sourceGLM-560.8%Kimi K2.559.1%GLM-5 leads this result
WideResearch
GLM-569.8%SourceKimi K2.572.7%SourceKimi K2.5 leads this result
CyberGym
GLM-543.2%SourceKimi K2.5—Not directly comparable
Gert Labs
Shared sourceGLM-550.99%Kimi K2.545.88%GLM-5 leads this result
BrowseComp
GLM-5—Kimi K2.560.6%SourceNot directly comparable
DeepSearchQA
GLM-5—Kimi K2.577.1%SourceNot directly comparable
ResearchClawBench
GLM-5—Kimi K2.514.0%SourceNot directly comparable
JobBench
GLM-5—Kimi K2.58.7%SourceNot directly comparable
Coding
SWE-bench Verified
GLM-577.8%SourceKimi K2.576.8%SourceGLM-5 leads this result
SWE-bench Verified*
Shared sourceGLM-572.8%Kimi K2.570.8%GLM-5 leads this result
SWE-bench Pro
GLM-555.1%SourceKimi K2.550.7%SourceGLM-5 leads this result
SWE Multilingual
GLM-573.3%SourceKimi K2.573%SourceGLM-5 leads this result
SWE-Rebench
Shared sourceGLM-562.8%Kimi K2.558.5%GLM-5 leads this result
React Native Evals
Shared sourceGLM-574.8%Kimi K2.577.2%Kimi K2.5 leads this result
LiveCodeBench v6
GLM-5—Kimi K2.585.0%SourceNot directly comparable
SciCode
GLM-5—Kimi K2.548.7%SourceNot directly comparable
Reasoning
Knowledge
GPQA
GLM-586%SourceKimi K2.587.6%SourceKimi K2.5 leads this result
GPQA-D
GLM-586.0%SourceKimi K2.587.6%SourceKimi K2.5 leads this result
SuperGPQA
Shared sourceGLM-566.8%Kimi K2.569.2%Kimi K2.5 leads this result
MMLU-Pro
GLM-585.7%SourceKimi K2.587.1%SourceKimi K2.5 leads this result
MMLU-Pro (Arcee)
Shared sourceGLM-585.8%Kimi K2.587.1%Kimi K2.5 leads this result
HLE
GLM-550.4%SourceKimi K2.530.1%SourceGLM-5 leads this result
Math
AIME26
GLM-595.8%SourceKimi K2.595.8%SourceTie
AIME25 (Arcee)
Shared sourceGLM-593.3%Kimi K2.596.3%Kimi K2.5 leads this result
HMMT Feb 2025
GLM-597.5%SourceKimi K2.595.4%SourceGLM-5 leads this result
HMMT Nov 2025
GLM-596.9%SourceKimi K2.591.1%SourceGLM-5 leads this result
HMMT Feb 2026
GLM-586.4%SourceKimi K2.587.1%SourceKimi K2.5 leads this result
MMAnswerBench
GLM-582.5%SourceKimi K2.581.8%SourceGLM-5 leads this result
FrontierMath v2 (Tiers 1-3)
Shared sourceGLM-516.434%Kimi K2.527.900%Kimi K2.5 leads this result
FrontierMath v2 (Tier 4)
Shared sourceGLM-52.100%Kimi K2.54.200%Kimi K2.5 leads this result
AIME 2025
GLM-5—Kimi K2.596.1%SourceNot directly comparable
Multilingual
MMLU-ProX
Shared sourceGLM-583.1%Kimi K2.582.3%GLM-5 leads this result
NOVA-63
Shared sourceGLM-555.1%Kimi K2.556.0%Kimi K2.5 leads this result
Multimodal
Instruction following
IFEval
Shared sourceGLM-592.6%Kimi K2.593.9%Kimi K2.5 leads this result
Frequently asked questions
Which is better, GLM-5 or Kimi K2.5?
GLM-5 has the higher public score estimate, 65.68 versus 58.81, but the 90% score intervals overlap. The higher estimate is not a decisive winner because the uncertainty ranges overlap.
Which is better for coding, GLM-5 or Kimi K2.5?
The current coding averages use different weighted benchmark sets, so BenchLM does not name a winner from them. Read the shared benchmark rows directly and test the models on the same task set.
Which is better for agentic tasks, GLM-5 or Kimi K2.5?
The current agentic tasks averages use different weighted benchmark sets, so BenchLM does not name a winner from them. Read the shared benchmark rows directly and test the models on the same task set.
Which costs less, GLM-5 or Kimi K2.5?
For the stated presets, chat costs $0.0026 on GLM-5 and $0.0021 on Kimi K2.5; repository review costs $0.0596 and $0.039; the cache-heavy agent loop costs $0.252 and $0.162. GLM-5 does not fit this workload in one request. GLM-5 has no published cached-input rate, so cached tokens use its listed input rate. Kimi K2.5 has no published cached-input rate, so cached tokens use its listed input rate.
Which has the larger context window, GLM-5 or Kimi K2.5?
Kimi K2.5 has the larger documented context window: 256K, compared with 200K.
Related comparisons
Last updated September 3, 2026
Watch GLM-5 vs Kimi K2.5
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 2,000+ readers.