Agentic
Like-for-like- GPT-5.6 Luna
- 84.1
- Kimi K2.5
- 55.0
- Weighted basis
- 2 vs 2 rows
- Reading
- GPT-5.6 Luna leads
Model comparison
GPT-5.6 Luna vs Kimi K2.5
Updated August 10, 2026. Public scores include evidence status and uncertainty. They are not guarantees for a specific workload.
GPT-5.6 Luna has the higher public score estimate, 66.87 versus 58.81, but the 90% score intervals overlap. Treat that as a lead, not a settled winner.
9 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.
Agentic work
Tool use, computer use, and multi-step task completion
GPT-5.6 Luna
GPT-5.6 Luna leads on the same 2 weighted benchmark rows.
Confidence: limited
Long documents
Prompts that approach the documented context limit
GPT-5.6 Luna
GPT-5.6 Luna has the larger documented context window.
Confidence: documented
Chat turn cost
1K fresh input + 500 output tokens
GPT-5.6 Luna
GPT-5.6 Luna 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
Cache-heavy agent loop cost
200K cached + 20K fresh input + 10K output tokens
GPT-5.6 Luna
GPT-5.6 Luna has the lower estimated token cost for this stated workload. Kimi K2.5 has no published cached-input rate, so cached tokens use its listed input rate.
Confidence: rate-fallback
Repository review cost
50K fresh input + 3K output tokens
GPT-5.6 Luna
GPT-5.6 Luna 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
What is actually comparable
Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.
13 GPT-5.6 Luna only
9 Shared
36 Kimi K2.5 only
- Shared results
- 9
- GPT-5.6 Luna only
- 13
- Kimi K2.5 only
- 36
- Like-for-like categories
- 2 / 8
3 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.
Multimodal
Like-for-like- GPT-5.6 Luna
- 78.4
- Kimi K2.5
- 78.5
- Weighted basis
- 1 vs 1 rows
- Reading
- Kimi K2.5 leads
Coding
Directional only- GPT-5.6 Luna
- 62.7
- Kimi K2.5
- 59.4
- Weighted basis
- 1 vs 4 rows
- Reading
- Directional only
Knowledge
Directional only- GPT-5.6 Luna
- 92.3
- Kimi K2.5
- 56.9
- Weighted basis
- 1 vs 4 rows
- Reading
- Directional only
Math
Directional only- GPT-5.6 Luna
- 73.6
- Kimi K2.5
- 60.6
- Weighted basis
- 2 vs 4 rows
- Reading
- Directional only
Reasoning
Not comparable- GPT-5.6 Luna
- 59.5
- Kimi K2.5
- 61.0
- Weighted basis
- 1 vs 1 rows
- Reading
- Not comparable
Multilingual
Not comparable- GPT-5.6 Luna
- Not measured
- Kimi K2.5
- 82.3
- Weighted basis
- 0 vs 1 rows
- Reading
- Not comparable
Instruction following
Not comparable- GPT-5.6 Luna
- Not measured
- Kimi K2.5
- 93.9
- Weighted basis
- 0 vs 1 rows
- Reading
- Not comparable
| Category | GPT-5.6 Luna | Kimi K2.5 | Weighted basis | Reading |
|---|---|---|---|---|
| Agentic | 84.1 | 55.0 | Like-for-like2 vs 2 rows | GPT-5.6 Luna leads |
| Multimodal | 78.4 | 78.5 | Like-for-like1 vs 1 rows | Kimi K2.5 leads |
| Coding | 62.7 | 59.4 | Directional only1 vs 4 rows | Directional only |
| Knowledge | 92.3 | 56.9 | Directional only1 vs 4 rows | Directional only |
| Math | 73.6 | 60.6 | Directional only2 vs 4 rows | Directional only |
| Reasoning | 59.5 | 61.0 | Not comparable1 vs 1 rows | Not comparable |
| Multilingual | Not measured | 82.3 | Not comparable0 vs 1 rows | Not comparable |
| Instruction following | Not measured | 93.9 | 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.
FrontierMath v2 (Tier 4)
Math
FrontierMath v2 (Tiers 1-3)
Math
Terminal-Bench 2.0
Agentic
BrowseComp
Agentic
SWE-bench Pro
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
- GPT-5.6 Luna
- $0.0008
- Fits in one request
- Kimi K2.5
- $0.0021
- Fits in one request
GPT-5.6 Luna has the lower modeled cost
Costs use the listed standard API rates.
Repository review
50K fresh input + 3K output tokens
- GPT-5.6 Luna
- $0.0136
- Fits in one request
- Kimi K2.5
- $0.039
- Fits in one request
GPT-5.6 Luna has the lower modeled cost
Costs use the listed standard API rates.
Cache-heavy agent loop
200K cached + 20K fresh input + 10K output tokens
- GPT-5.6 Luna
- $0.02
- Fits in one request
- Kimi K2.5
- $0.162
- Fits in one request
- Cached input priced at the published list-input rate
GPT-5.6 Luna has the lower modeled cost
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.
SpecificationGPT-5.6 LunaKimi K2.5
Context window
Maximum documented context; output-token limits may be lower.
GPT-5.6 Luna
1.05M
OpenAI model catalogKimi K2.5
256K
API model ID
GPT-5.6 Luna
gpt-5.6-luna
OpenAI model catalogKimi 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.
GPT-5.6 Luna
$0.02 per 1M cached input tokens
OpenAI pricingKimi K2.5
Not published
Documented inputs
GPT-5.6 Luna
text, image
OpenAI model catalogKimi K2.5
Not sourced
Documented outputs
GPT-5.6 Luna
Kimi K2.5
Not sourced
Provider availability
GPT-5.6 Luna
Generally Available · OpenAI Responses API
OpenAI model catalogKimi K2.5
Not sourced
Reasoning profile
GPT-5.6 Luna
Reasoning
Kimi K2.5
Non-Reasoning
Weight access
GPT-5.6 Luna
Proprietary
Kimi K2.5
Open Weight
License
GPT-5.6 Luna
Proprietary
Kimi K2.5
Open Weight
Release date
GPT-5.6 Luna
2026-07-09
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
- GPT-5.6 Luna has the higher public score estimate, 66.87 versus 58.81, but the 90% score intervals overlap.
- Workload cost
- Repository review: $0.0136 vs $0.039. Cache-heavy agent loop: $0.02 vs $0.162.
- Context tradeoff
- GPT-5.6 Luna has the larger documented window (1.05M).
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.
GPT-5.6 Luna
API / mo$1,050
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 evidence58 rows
Agentic
Terminal-Bench 2.0
GPT-5.6 Luna84.7%SourceKimi K2.550.8%SourceGPT-5.6 Luna leads this result
BrowseComp
GPT-5.6 Luna83.3%SourceKimi K2.560.6%SourceGPT-5.6 Luna leads this result
OSWorld 2.0
GPT-5.6 Luna45.6%SourceKimi K2.5—Not directly comparable
CyberGym
GPT-5.6 Luna77.9%SourceKimi K2.5—Not directly comparable
ExploitGym
GPT-5.6 Luna12.4%SourceKimi K2.5—Not directly comparable
Toolathlon
GPT-5.6 Luna53.4%SourceKimi K2.527.8%SourceGPT-5.6 Luna leads this result
Claw-Eval
GPT-5.6 Luna—Kimi K2.552.3%SourceNot directly comparable
QwenClawBench
GPT-5.6 Luna—Kimi K2.554.3%SourceNot directly comparable
τ³-bench results
GPT-5.6 Luna—Kimi K2.565.7%SourceNot directly comparable
DeepSearchQA
GPT-5.6 Luna—Kimi K2.577.1%SourceNot directly comparable
DeepPlanning
GPT-5.6 Luna—Kimi K2.514.4%SourceNot directly comparable
MCP Atlas
GPT-5.6 Luna—Kimi K2.529.5%SourceNot directly comparable
MCP-Tasks
GPT-5.6 Luna—Kimi K2.559.1%SourceNot directly comparable
WideResearch
GPT-5.6 Luna—Kimi K2.572.7%SourceNot directly comparable
Gert Labs
GPT-5.6 Luna—Kimi K2.545.88%SourceNot directly comparable
ResearchClawBench
GPT-5.6 Luna—Kimi K2.514.0%SourceNot directly comparable
JobBench
GPT-5.6 Luna—Kimi K2.58.7%SourceNot directly comparable
Coding
SWE-bench Pro
GPT-5.6 Luna62.7%SourceKimi K2.550.7%SourceGPT-5.6 Luna leads this result
Terminal-Bench 2.0
GPT-5.6 Luna84.7%SourceKimi K2.5—Not directly comparable
deepSwe
GPT-5.6 Luna67.2%SourceKimi K2.5—Not directly comparable
FrontierCode 1.1 Extended
GPT-5.6 Luna55.1%SourceKimi K2.5—Not directly comparable
cursorBench32
GPT-5.6 Luna61.1%SourceKimi K2.5—Not directly comparable
SWE-bench Verified
GPT-5.6 Luna—Kimi K2.576.8%SourceNot directly comparable
SWE-bench Verified*
GPT-5.6 Luna—Kimi K2.570.8%SourceNot directly comparable
LiveCodeBench v6
GPT-5.6 Luna—Kimi K2.585.0%SourceNot directly comparable
SWE Multilingual
GPT-5.6 Luna—Kimi K2.573%SourceNot directly comparable
SWE-Rebench
GPT-5.6 Luna—Kimi K2.558.5%SourceNot directly comparable
React Native Evals
GPT-5.6 Luna—Kimi K2.577.2%SourceNot directly comparable
SciCode
GPT-5.6 Luna—Kimi K2.548.7%SourceNot directly comparable
Reasoning
Knowledge
GPQA
GPT-5.6 Luna92.3%SourceKimi K2.587.6%SourceGPT-5.6 Luna leads this result
GPQA-D
GPT-5.6 Luna92.3%SourceKimi K2.587.6%SourceGPT-5.6 Luna leads this result
HealthBench Professional
GPT-5.6 Luna55.7%SourceKimi K2.5—Not directly comparable
HealthBench Hard
GPT-5.6 Luna32.0%SourceKimi K2.5—Not directly comparable
SuperGPQA
GPT-5.6 Luna—Kimi K2.569.2%SourceNot directly comparable
MMLU-Pro
GPT-5.6 Luna—Kimi K2.587.1%SourceNot directly comparable
MMLU-Pro (Arcee)
GPT-5.6 Luna—Kimi K2.587.1%SourceNot directly comparable
HLE
GPT-5.6 Luna—Kimi K2.530.1%SourceNot directly comparable
Math
FrontierMath (legacy)
GPT-5.6 Luna78.6%SourceKimi K2.5—Not directly comparable
FrontierMath v2 (Tiers 1-3)
GPT-5.6 Luna78.600%SourceKimi K2.527.900%SourceGPT-5.6 Luna leads this result
FrontierMath v2 (Tier 4)
GPT-5.6 Luna58.500%SourceKimi K2.54.200%SourceGPT-5.6 Luna leads this result
AIME 2025
GPT-5.6 Luna—Kimi K2.596.1%SourceNot directly comparable
AIME26
GPT-5.6 Luna—Kimi K2.595.8%SourceNot directly comparable
AIME25 (Arcee)
GPT-5.6 Luna—Kimi K2.596.3%SourceNot directly comparable
HMMT Feb 2025
GPT-5.6 Luna—Kimi K2.595.4%SourceNot directly comparable
HMMT Nov 2025
GPT-5.6 Luna—Kimi K2.591.1%SourceNot directly comparable
HMMT Feb 2026
GPT-5.6 Luna—Kimi K2.587.1%SourceNot directly comparable
MMAnswerBench
GPT-5.6 Luna—Kimi K2.581.8%SourceNot directly comparable
Multilingual
Multimodal
MMMU-Pro
GPT-5.6 Luna78.4%SourceKimi K2.578.5%SourceKimi K2.5 leads this result
MMMU-Pro w/ Python
GPT-5.6 Luna79.5%SourceKimi K2.5—Not directly comparable
Video-MME
GPT-5.6 Luna—Kimi K2.587.4%SourceNot directly comparable
MMVU
GPT-5.6 Luna—Kimi K2.580.4%SourceNot directly comparable
VideoMMMU
GPT-5.6 Luna—Kimi K2.586.6%SourceNot directly comparable
Instruction following
IFEval
GPT-5.6 Luna—Kimi K2.593.9%SourceNot directly comparable
Frequently asked questions
Which is better, GPT-5.6 Luna or Kimi K2.5?
GPT-5.6 Luna has the higher public score estimate, 66.87 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, GPT-5.6 Luna 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, GPT-5.6 Luna or Kimi K2.5?
GPT-5.6 Luna leads the like-for-like agentic tasks comparison across 2 shared weighted benchmark rows.
Which costs less, GPT-5.6 Luna or Kimi K2.5?
For the stated presets, chat costs $0.0008 on GPT-5.6 Luna and $0.0021 on Kimi K2.5; repository review costs $0.0136 and $0.039; the cache-heavy agent loop costs $0.02 and $0.162. Kimi K2.5 has no published cached-input rate, so cached tokens use its listed input rate.
Which has the larger context window, GPT-5.6 Luna or Kimi K2.5?
GPT-5.6 Luna has the larger documented context window: 1.05M, compared with 256K.
Related comparisons
Last updated August 10, 2026
Watch GPT-5.6 Luna 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.