Agentic
Like-for-like- GPT-5.5
- 61.0
- Supported · #17/152
- Kimi K3
- 71.9
- Supported · #4/152
- Basis
- BenchAlign lane · 14 vs 12 public rows
- Reading
- Kimi K3 leads
GPT-5.5 vs Kimi K3
Updated September 10, 2026. Public scores include evidence status and uncertainty. They are not guarantees for a specific workload.
Decision reading
Kimi K3 has the higher public score estimate, 74.82 versus 72.07, but the 90% score intervals overlap. Treat that as a lead, not a settled winner.
18 results are shared. Category rows resting on Estimated evidence or different benchmark sets are marked directional and do not name a winner.
Share or export
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
Kimi K3
Kimi K3 leads on the public agentic lane, 71.9 to 61, with Supported evidence for both models and non-overlapping 90% intervals.
Confidence: stronger
Long documents
Prompts that approach the documented context limit
Kimi K3
Kimi K3 has the larger documented context window.
Confidence: documented
Chat turn cost
1K fresh input + 500 output tokens
Kimi K3
Kimi K3 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
Kimi K3
Kimi K3 has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates
Repository review cost
50K fresh input + 3K output tokens
Kimi K3
Kimi K3 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
No clear pick
The like-for-like coding result is a practical tie on the public lane (within 0.5 points).
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.
21 GPT-5.5 only
18 Shared
26 Kimi K3 only
- Shared results
- 18
- GPT-5.5 only
- 21
- Kimi K3 only
- 26
- Like-for-like categories
- 3 / 8
2 categories rest on Estimated evidence or different evidence sets. Those rows remain visible for coverage context but do not name a winner.
Category results, on a stated basis
Each row shows the public-lane category score for both models: the BenchAlign lane for agentic, coding, and knowledge, and the weighted public rows elsewhere, with the same rank each model page reports. A row is like-for-like only when both scores rest on Supported evidence or the same weighted set. Directional and not-comparable rows remain visible, but they never receive a winner.
Coding
Like-for-like- GPT-5.5
- 67.6
- Supported · #7/151
- Kimi K3
- 68.0
- Supported · #6/151
- Basis
- BenchAlign lane · 9 vs 13 public rows
- Reading
- Practical tie
Knowledge
Like-for-like- GPT-5.5
- 72.9
- Supported · #7/183
- Kimi K3
- 72.0
- Supported · #8/183
- Basis
- BenchAlign lane · 6 vs 6 public rows
- Reading
- GPT-5.5 leads · intervals overlap
Reasoning
Directional only- GPT-5.5
- 64.0
- #13/20
- Kimi K3
- 78.5
- #3/20
- Basis
- Provisional lane · 2 vs 0 weighted rows
- Reading
- Directional only
Multimodal
Directional only- GPT-5.5
- 71.3
- #19/48
- Kimi K3
- 89.5
- #1/48
- Basis
- Provisional lane · 2 vs 3 weighted rows
- Reading
- Directional only
Math
Not comparable- GPT-5.5
- 69.6
- Unranked · 3 rankable rows
- Kimi K3
- Not ranked
- Basis
- Provisional lane · 2 vs 0 weighted rows
- Reading
- Not comparable
Multilingual
Not comparable- GPT-5.5
- Not ranked
- Kimi K3
- Not ranked
- Basis
- Provisional lane · 0 vs 0 weighted rows
- Reading
- Not comparable
Instruction following
Not comparable- GPT-5.5
- 93.2
- #7/123
- Kimi K3
- Not ranked
- Basis
- Provisional lane · 0 vs 0 weighted rows
- Reading
- Not comparable
| Category | GPT-5.5 | Kimi K3 | Basis | Reading |
|---|---|---|---|---|
| Agentic | 61.0Supported · #17/152 | 71.9Supported · #4/152 | Like-for-likeBenchAlign lane · 14 vs 12 public rows | Kimi K3 leads |
| Coding | 67.6Supported · #7/151 | 68.0Supported · #6/151 | Like-for-likeBenchAlign lane · 9 vs 13 public rows | Practical tie |
| Knowledge | 72.9Supported · #7/183 | 72.0Supported · #8/183 | Like-for-likeBenchAlign lane · 6 vs 6 public rows | GPT-5.5 leads · intervals overlap |
| Reasoning | 64.0#13/20 | 78.5#3/20 | Directional onlyProvisional lane · 2 vs 0 weighted rows | Directional only |
| Multimodal | 71.3#19/48 | 89.5#1/48 | Directional onlyProvisional lane · 2 vs 3 weighted rows | Directional only |
| Math | 69.6Unranked · 3 rankable rows | Not ranked | Not comparableProvisional lane · 2 vs 0 weighted rows | Not comparable |
| Multilingual | Not ranked | Not ranked | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
| Instruction following | 93.2#7/123 | Not ranked | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
Ranks count the models scored in each category’s lane, so the agentic, coding, and knowledge denominators (BenchAlign) differ from the provisional-lane categories. Unranked scores sit on the lane’s scale but never name a winner. Like-for-like rows within 0.5 points read as a practical tie.
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.
OfficeQA Pro
Multimodal
BrowseComp
Agentic
Terminal-Bench 2.0
Agentic
HLE
Knowledge
cursorBench32
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-5.5
- $0.02
- Fits in one request
- Kimi K3
- $0.0105
- Fits in one request
Kimi K3 has the lower modeled cost
Costs use the listed standard API rates.
Repository review
50K fresh input + 3K output tokens
- GPT-5.5
- $0.34
- Fits in one request
- Kimi K3
- $0.195
- Fits in one request
Kimi K3 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-5.5
- $0.5
- Fits in one request
- Kimi K3
- $0.27
- Fits in one request
Kimi K3 has the lower modeled cost
Costs use the listed standard API rates.
Specification differences
Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.
SpecificationGPT-5.5Kimi K3
Context window
Maximum documented context; output-token limits may be lower.
GPT-5.5
Kimi K3
1.05M
API model ID
GPT-5.5
gpt-5.5
OpenAI pricingKimi K3
Not sourced
Cached-input rate
A missing cached-input rate falls back to the listed input rate only in the stated workload estimate.
GPT-5.5
$0.5 per 1M cached input tokens
OpenAI pricingKimi K3
$0.3 per 1M cached input tokens
Documented inputs
GPT-5.5
Not sourced
Kimi K3
Not sourced
Documented outputs
GPT-5.5
Not sourced
Kimi K3
Not sourced
Provider availability
GPT-5.5
Not sourced
Kimi K3
Not sourced
Reasoning profile
GPT-5.5
Reasoning
Kimi K3
Reasoning
Weight access
GPT-5.5
Proprietary
Kimi K3
Pending
License
GPT-5.5
Proprietary
Kimi K3
Pending
Release date
GPT-5.5
2026-04-23
Kimi K3
2026-07-16
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
- Kimi K3 has the higher public score estimate, 74.82 versus 72.07, but the 90% score intervals overlap.
- Workload cost
- Repository review: $0.34 vs $0.195. Cache-heavy agent loop: $0.5 vs $0.27.
- Context tradeoff
- Kimi K3 has the larger documented window (1.05M).
Run the same representative tasks against both endpoints before changing production traffic.
Benchmark evidence
The full public result ledger is available for audit without forcing a wide desktop table onto a phone.
Browse raw public benchmark evidence65 rows
Agentic
Terminal-Bench 2.0
GPT-5.582%SourceKimi K388.3%SourceKimi K3 leads this result
CyberGym
GPT-5.581.8%SourceKimi K3—Not directly comparable
BrowseComp
GPT-5.584.4%SourceKimi K391.2%SourceKimi K3 leads this result
OSWorld-Verified
GPT-5.578.7%SourceKimi K3—Not directly comparable
MCP Atlas
GPT-5.575.3%SourceKimi K384.2%SourceKimi K3 leads this result
Toolathlon
GPT-5.555.6%SourceKimi K3—Not directly comparable
τ²-bench results
GPT-5.598%SourceKimi K3—Not directly comparable
Gert Labs
GPT-5.572.93%SourceKimi K3—Not directly comparable
ResearchClawBench
GPT-5.517.0%SourceKimi K3—Not directly comparable
OSWorld 2.0
GPT-5.513.0%SourceKimi K3—Not directly comparable
JobBench
GPT-5.542.7%SourceKimi K352.9%SourceKimi K3 leads this result
ExploitGym
GPT-5.513.4%SourceKimi K3—Not directly comparable
Terminal-Bench 2.1 (Vals)
GPT-5.576.4%SourceKimi K380.9%SourceKimi K3 leads this result
ApprenticeBench
Shared sourceGPT-5.520%Kimi K318%GPT-5.5 leads this result
DeepSearchQA
GPT-5.5—Kimi K395.0%SourceNot directly comparable
Toolathlon-Verified
GPT-5.5—Kimi K373.2%SourceNot directly comparable
AutomationBench
GPT-5.5—Kimi K330.8%SourceNot directly comparable
APEX-Agents
GPT-5.5—Kimi K337.6%SourceNot directly comparable
SpreadsheetBench 2
GPT-5.5—Kimi K334.8%SourceNot directly comparable
DECK-Bench
GPT-5.5—Kimi K373.5%SourceNot directly comparable
Coding
SWE-bench Pro
GPT-5.558.6%SourceKimi K3—Not directly comparable
Terminal-Bench 2.0
GPT-5.582.0%SourceKimi K3—Not directly comparable
Vibe Code Bench
GPT-5.569.85%SourceKimi K3—Not directly comparable
React Native Evals
GPT-5.584.7%SourceKimi K3—Not directly comparable
cursorBench31
GPT-5.559.2%SourceKimi K3—Not directly comparable
cursorBench32
Shared sourceGPT-5.558.4%Kimi K360.8%Kimi K3 leads this result
FrontierCode 1.1 Main
GPT-5.543.0%SourceKimi K3—Not directly comparable
LiveCodeBench (Vals)
GPT-5.585.3%SourceKimi K387.2%SourceKimi K3 leads this result
SWE-bench (Vals)
GPT-5.582.6%SourceKimi K393.4%SourceKimi K3 leads this result
DeepSWE
GPT-5.5—Kimi K367.5%SourceNot directly comparable
FrontierSWE
GPT-5.5—Kimi K381.2%SourceNot directly comparable
ProgramBench
GPT-5.5—Kimi K377.8%SourceNot directly comparable
Kimi Code Bench v2
GPT-5.5—Kimi K372.9%SourceNot directly comparable
sweMarathon
GPT-5.5—Kimi K342%SourceNot directly comparable
PostTrain Bench
GPT-5.5—Kimi K336.6%SourceNot directly comparable
MLS-Bench Lite
GPT-5.5—Kimi K348.3%SourceNot directly comparable
VulcanBench v3
GPT-5.5—Kimi K373.7%SourceNot directly comparable
OpenHarmony Bench
GPT-5.5—Kimi K357.3%SourceNot directly comparable
FrontierSWE v2
GPT-5.5—Kimi K325.9%SourceNot directly comparable
Reasoning
Knowledge
GPQA
GPT-5.593.6%SourceKimi K393.5%SourceGPT-5.5 leads this result
GPQA-D
GPT-5.593.6%SourceKimi K393.5%SourceGPT-5.5 leads this result
HLE
GPT-5.552.2%SourceKimi K356%SourceKimi K3 leads this result
HLE w/o tools
GPT-5.541.4%SourceKimi K343.5%SourceKimi K3 leads this result
GPQA Diamond (Vals)
GPT-5.593.2%SourceKimi K392.9%SourceGPT-5.5 leads this result
MMLU-Pro (Vals)
GPT-5.588.1%SourceKimi K388.0%SourceGPT-5.5 leads this result
Math
Multimodal
MMMU-Pro
GPT-5.581.2%SourceKimi K381.6%SourceKimi K3 leads this result
MMMU-Pro w/ Python
GPT-5.583.2%SourceKimi K383.4%SourceKimi K3 leads this result
OfficeQA Pro
GPT-5.554.1%SourceKimi K363.3%SourceKimi K3 leads this result
CharXiv w/o tools
GPT-5.5—Kimi K384.8%SourceNot directly comparable
CharXiv
GPT-5.5—Kimi K391.3%SourceNot directly comparable
MathVision
GPT-5.5—Kimi K394.3%SourceNot directly comparable
MathVision w/ Python
GPT-5.5—Kimi K397.8%SourceNot directly comparable
BabyVision w/ Python
GPT-5.5—Kimi K385.7%SourceNot directly comparable
ZeroBench
GPT-5.5—Kimi K323.0%SourceNot directly comparable
ZeroBench w/ Python
GPT-5.5—Kimi K341.0%SourceNot directly comparable
WorldVQA ForceAnswer
GPT-5.5—Kimi K351.0%SourceNot directly comparable
OmniDocBench
GPT-5.5—Kimi K391.1%SourceNot directly comparable
PerceptionBench
GPT-5.5—Kimi K358.5%SourceNot directly comparable
Frequently asked questions
Which is better, GPT-5.5 or Kimi K3?
Kimi K3 has the higher public score estimate, 74.82 versus 72.07, 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-5.5 or Kimi K3?
The like-for-like coding row is a practical tie on the public lane, 67.6 against 68, inside the 0.5-point band BenchLM treats as level.
Which is better for agentic tasks, GPT-5.5 or Kimi K3?
Kimi K3 leads the public agentic tasks lane, 71.9 to 61, with Supported evidence for both models and non-overlapping 90% intervals.
Which costs less, GPT-5.5 or Kimi K3?
For the stated presets, chat costs $0.02 on GPT-5.5 and $0.0105 on Kimi K3; repository review costs $0.34 and $0.195; the cache-heavy agent loop costs $0.5 and $0.27. Costs use the listed standard API rates.
Which has the larger context window, GPT-5.5 or Kimi K3?
Kimi K3 has the larger documented context window: 1.05M, compared with 1M.
Related comparisons
Last updated September 10, 2026
Watch GPT-5.5 vs Kimi K3
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 2,000+ readers.