Knowledge
Directional only- Kimi K3
- 61.0
- Qwen3 235B 2507
- 78.9
- Weighted basis
- 2 vs 3 rows
- Reading
- Directional only
Kimi K3 vs Qwen3 235B 2507
Updated August 19, 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, 80.53 versus 56.44, and the 90% score intervals do not overlap.
1 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 K3
Kimi K3 has the larger documented context window.
Confidence: documented
Show secondary and unsupported calls
Coding work
Code generation, repair, and software-engineering tasks
Not enough matched evidence
No shared weighted benchmark basis supports a winner.
Confidence: limited
Agentic work
Tool use, computer use, and multi-step task completion
Not enough matched evidence
No shared weighted benchmark basis supports a winner.
Confidence: limited
Chat turn cost
1K fresh input + 500 output tokens
Not enough matched evidence
A complete comparable API-rate estimate is not available for both models.
Confidence: listed-rates
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. Qwen3 235B 2507 does not fit this workload in one request. Qwen3 235B 2507 has no comparable published API token rate.
Confidence: listed-rates
Repository review cost
50K fresh input + 3K output tokens
Not enough matched evidence
A complete comparable API-rate estimate is not available for both models.
Confidence: listed-rates
What is actually comparable
Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.
39 Kimi K3 only
1 Shared
3 Qwen3 235B 2507 only
- Shared results
- 1
- Kimi K3 only
- 39
- Qwen3 235B 2507 only
- 3
- Like-for-like categories
- 0 / 8
1 category uses 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.
Agentic
Not comparable- Kimi K3
- 89.5
- Qwen3 235B 2507
- Not measured
- Weighted basis
- 2 vs 0 rows
- Reading
- Not comparable
Coding
Not comparable- Kimi K3
- Not measured
- Qwen3 235B 2507
- Not measured
- Weighted basis
- 0 vs 0 rows
- Reading
- Not comparable
Reasoning
Not comparable- Kimi K3
- Not measured
- Qwen3 235B 2507
- Not measured
- Weighted basis
- 0 vs 0 rows
- Reading
- Not comparable
Math
Not comparable- Kimi K3
- Not measured
- Qwen3 235B 2507
- Not measured
- Weighted basis
- 0 vs 0 rows
- Reading
- Not comparable
Multilingual
Not comparable- Kimi K3
- Not measured
- Qwen3 235B 2507
- 79.4
- Weighted basis
- 0 vs 1 rows
- Reading
- Not comparable
Multimodal
Not comparable- Kimi K3
- 78.5
- Qwen3 235B 2507
- Not measured
- Weighted basis
- 3 vs 0 rows
- Reading
- Not comparable
Instruction following
Not comparable- Kimi K3
- Not measured
- Qwen3 235B 2507
- Not measured
- Weighted basis
- 0 vs 0 rows
- Reading
- Not comparable
| Category | Kimi K3 | Qwen3 235B 2507 | Weighted basis | Reading |
|---|---|---|---|---|
| Knowledge | 61.0 | 78.9 | Directional only2 vs 3 rows | Directional only |
| Agentic | 89.5 | Not measured | Not comparable2 vs 0 rows | Not comparable |
| Coding | Not measured | Not measured | Not comparable0 vs 0 rows | Not comparable |
| Reasoning | Not measured | Not measured | Not comparable0 vs 0 rows | Not comparable |
| Math | Not measured | Not measured | Not comparable0 vs 0 rows | Not comparable |
| Multilingual | Not measured | 79.4 | Not comparable0 vs 1 rows | Not comparable |
| Multimodal | 78.5 | Not measured | Not comparable3 vs 0 rows | Not comparable |
| Instruction following | Not measured | Not measured | Not comparable0 vs 0 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.
GPQA
Knowledge
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
- Kimi K3
- $0.0105
- Fits in one request
- Qwen3 235B 2507
- Self-hosted; infrastructure cost varies
- Fits in one request
Qwen3 235B 2507 has no comparable published API token rate.
Repository review
50K fresh input + 3K output tokens
- Kimi K3
- $0.195
- Fits in one request
- Qwen3 235B 2507
- Self-hosted; infrastructure cost varies
- Fits in one request
Qwen3 235B 2507 has no comparable published API token rate.
Cache-heavy agent loop
200K cached + 20K fresh input + 10K output tokens
- Kimi K3
- $0.27
- Fits in one request
- Qwen3 235B 2507
- Self-hosted; infrastructure cost varies
- Does not fit in one request
- Cached-input rate unavailable
Qwen3 235B 2507 does not fit this workload in one request. Qwen3 235B 2507 has no comparable published API token rate.
Specification differences
Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.
SpecificationKimi K3Qwen3 235B 2507
Context window
Maximum documented context; output-token limits may be lower.
Kimi K3
1.05M
Qwen3 235B 2507
128K
API model ID
Kimi K3
Not sourced
Qwen3 235B 2507
Not sourced
Cached-input rate
A missing cached-input rate falls back to the listed input rate only in the stated workload estimate.
Kimi K3
$0.3 per 1M cached input tokens
Qwen3 235B 2507
No comparable hosted API rate
Documented inputs
Kimi K3
Not sourced
Qwen3 235B 2507
Not sourced
Documented outputs
Kimi K3
Not sourced
Qwen3 235B 2507
Not sourced
Provider availability
Kimi K3
Not sourced
Qwen3 235B 2507
Not sourced
Reasoning profile
Kimi K3
Reasoning
Qwen3 235B 2507
Non-Reasoning
Weight access
Kimi K3
Pending
Qwen3 235B 2507
Open Weight
License
Kimi K3
Pending
Qwen3 235B 2507
Open Weight
Release date
Kimi K3
2026-07-16
Qwen3 235B 2507
2025-07-01
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
- Kimi K3 has the higher public score, 80.53 versus 56.44, and the 90% score intervals do not overlap.
- Workload cost
- A complete comparable API-rate estimate is not available for both models.
- 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 evidence43 rows
Agentic
Terminal-Bench 2.0
Kimi K388.3%SourceQwen3 235B 2507—Not directly comparable
BrowseComp
Kimi K391.2%SourceQwen3 235B 2507—Not directly comparable
DeepSearchQA
Kimi K395.0%SourceQwen3 235B 2507—Not directly comparable
Toolathlon-Verified
Kimi K373.2%SourceQwen3 235B 2507—Not directly comparable
MCP Atlas
Kimi K384.2%SourceQwen3 235B 2507—Not directly comparable
AutomationBench
Kimi K330.8%SourceQwen3 235B 2507—Not directly comparable
JobBench
Kimi K352.9%SourceQwen3 235B 2507—Not directly comparable
APEX-Agents
Kimi K337.6%SourceQwen3 235B 2507—Not directly comparable
SpreadsheetBench 2
Kimi K334.8%SourceQwen3 235B 2507—Not directly comparable
DECK-Bench
Kimi K373.5%SourceQwen3 235B 2507—Not directly comparable
Coding
deepSwe
Kimi K367.5%SourceQwen3 235B 2507—Not directly comparable
cursorBench32
Kimi K360.8%SourceQwen3 235B 2507—Not directly comparable
FrontierSWE
Kimi K381.2%SourceQwen3 235B 2507—Not directly comparable
ProgramBench
Kimi K377.8%SourceQwen3 235B 2507—Not directly comparable
Kimi Code Bench v2
Kimi K372.9%SourceQwen3 235B 2507—Not directly comparable
sweMarathon
Kimi K342%SourceQwen3 235B 2507—Not directly comparable
PostTrain Bench
Kimi K336.6%SourceQwen3 235B 2507—Not directly comparable
MLS-Bench Lite
Kimi K348.3%SourceQwen3 235B 2507—Not directly comparable
VulcanBench v3
Kimi K373.9%SourceQwen3 235B 2507—Not directly comparable
APEX-SWE
Kimi K348.0%SourceQwen3 235B 2507—Not directly comparable
EEBench
Kimi K338.3%SourceQwen3 235B 2507—Not directly comparable
InferenceEval
Kimi K339.8%SourceQwen3 235B 2507—Not directly comparable
KernelBench Internal
Kimi K368.5%SourceQwen3 235B 2507—Not directly comparable
Knowledge
GPQA
Kimi K393.5%SourceQwen3 235B 250777.5%SourceKimi K3 leads this result
GPQA-D
Kimi K393.5%SourceQwen3 235B 2507—Not directly comparable
HLE
Kimi K356%SourceQwen3 235B 2507—Not directly comparable
HLE w/o tools
Kimi K343.5%SourceQwen3 235B 2507—Not directly comparable
SuperGPQA
Kimi K3—Qwen3 235B 250762.6%SourceNot directly comparable
MMLU-Pro
Kimi K3—Qwen3 235B 250783%SourceNot directly comparable
Multilingual
MMLU-ProX
Kimi K3—Qwen3 235B 250779.4%SourceNot directly comparable
Multimodal
OfficeQA Pro
Kimi K363.3%SourceQwen3 235B 2507—Not directly comparable
MMMU-Pro
Kimi K381.6%SourceQwen3 235B 2507—Not directly comparable
MMMU-Pro w/ Python
Kimi K383.4%SourceQwen3 235B 2507—Not directly comparable
CharXiv w/o tools
Kimi K384.8%SourceQwen3 235B 2507—Not directly comparable
CharXiv
Kimi K391.3%SourceQwen3 235B 2507—Not directly comparable
MathVision
Kimi K394.3%SourceQwen3 235B 2507—Not directly comparable
MathVision w/ Python
Kimi K397.8%SourceQwen3 235B 2507—Not directly comparable
BabyVision w/ Python
Kimi K385.7%SourceQwen3 235B 2507—Not directly comparable
ZeroBench
Kimi K323.0%SourceQwen3 235B 2507—Not directly comparable
ZeroBench w/ Python
Kimi K341.0%SourceQwen3 235B 2507—Not directly comparable
WorldVQA ForceAnswer
Kimi K351.0%SourceQwen3 235B 2507—Not directly comparable
OmniDocBench
Kimi K391.1%SourceQwen3 235B 2507—Not directly comparable
PerceptionBench
Kimi K358.5%SourceQwen3 235B 2507—Not directly comparable
Frequently asked questions
Which is better, Kimi K3 or Qwen3 235B 2507?
Kimi K3 has the higher public score, 80.53 versus 56.44, and the 90% score intervals do not overlap. That is the clearest overall quality signal in the current public data.
Which is better for coding, Kimi K3 or Qwen3 235B 2507?
The published evidence does not provide a shared weighted coding basis for both models, so BenchLM does not name a coding winner.
Which is better for agentic tasks, Kimi K3 or Qwen3 235B 2507?
The published evidence does not provide a shared weighted agentic tasks basis for both models, so BenchLM does not name a agentic tasks winner.
Which costs less, Kimi K3 or Qwen3 235B 2507?
Both models do not have comparable published API token rates, so this page does not name a universal price winner.
Which has the larger context window, Kimi K3 or Qwen3 235B 2507?
Kimi K3 has the larger documented context window: 1.05M, compared with 128K.
Related comparisons
Last updated August 19, 2026
Watch Kimi K3 vs Qwen3 235B 2507
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 2,000+ readers.