Long documents
Prompts that approach the documented context limit
Kimi K2.5
Kimi K2.5 has the larger documented context window.
Confidence: documented
Kimi K2.5 vs Qwen3.5 397B
Updated October 17, 2026. Rank says Qwen3.5 397B is ahead. Price, access, and your workload can each overturn that. Public scores include evidence status and uncertainty.
Share or export
Decision reading
Qwen3.5 397B has the higher public point estimate, 55.34 versus 53.05. Their conditional score ranges overlap. These ranges do not establish rank confidence. 55 results are shared. Category rows resting on Estimated evidence or different benchmark sets are marked directional and do not name a winner.
- Shared results
- 55
- Kimi K2.5 only
- 13
- Qwen3.5 397B only
- 7
- Like-for-like categories
- 2 / 8
Supported: Kimi K2.5 · Estimated: Qwen3.5 397B. Conditional ranges do not establish rank confidence.How the comparison works
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.
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-ratesCoding work
Code generation, repair, and software-engineering tasks
Not enough matched evidence
Qwen3.5 397B is not ranked on the public lane for coding, so no winner is named for coding.
Confidence: limitedAgentic work
Tool use, computer use, and multi-step task completion
Not enough matched evidence
Kimi K2.5 and Qwen3.5 397B are scored on Estimated evidence for agentic, so the reading is directional rather than like-for-like.
Confidence: limitedCache-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. Qwen3.5 397B does not fit this workload in one request. Kimi K2.5 has no published cached-input rate, so cached tokens use its listed input rate. Qwen3.5 397B has no published cached-input rate, so cached tokens use its listed input rate.
Confidence: rate-fallback
Which one for a specific job
Choose a job from the LLM Selector's task catalog to see the category row it rests on, under the same basis rules as the table below. A directional row stays directional; choosing a job never creates a winner.
The same task catalog as the LLM Selector. Each job names the evidence surface it rests on; nothing here adds a new score.
38.5Kimi K2.5—Qwen3.5 397B
Not comparable · BenchAlign v5.8
The coding row is not comparable on the public lane: at least one model is not measured or not ranked there.
Coding scores combine specific tasks and setups. Match the editor, harness, and effort to your workflow.
Same basis rules as the category table below
What is actually comparable
Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.
13 Kimi K2.5 only
55 Shared
7 Qwen3.5 397B only
3 categories rest on Estimated evidence or different evidence sets. Those rows remain visible for coverage context but do not name a winner.
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.
BrowseCompAgentic
Normalized gap 8.4Kimi K2.5:60.6%Qwen3.5 397B:69%SWE MultilingualCoding
Normalized gap 3.7Kimi K2.5:73%Qwen3.5 397B:69.3%CharXivMultimodal
Normalized gap 3.3Kimi K2.5:77.5%Qwen3.5 397B:80.8%AIME26Math
Normalized gap 2.5Kimi K2.5:95.8%Qwen3.5 397B:93.3%MMLU-ProXMultilingual
Normalized gap 2.4Kimi K2.5:82.3%Qwen3.5 397B:84.7%
Bars run 0–100 on each benchmark’s normalized display scale
Category results, on a stated basis
Each row shows the public-lane category score for both models: the BenchAlign v5.8 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.
Multimodal
Like-for-like- Kimi K2.5
- 70.4
- #32/54
- Qwen3.5 397B
- 73.6
- #29/54
- Basis
- Provisional lane · 2 vs 2 weighted rows
- Reading
- Qwen3.5 397B leads
Multilingual
Like-for-like- Kimi K2.5
- 87.3
- #8/17
- Qwen3.5 397B
- 93.8
- #5/17
- Basis
- Provisional lane · 1 vs 1 weighted rows
- Reading
- Qwen3.5 397B leads
Agentic
Directional only- Kimi K2.5
- 40.7
- Estimated · #66/128
- Qwen3.5 397B
- 43.8
- Estimated · #61/128
- Basis
- BenchAlign v5.8 lane · 16 vs 14 public rows
- Reading
- Directional only
Knowledge
Directional only- Kimi K2.5
- 51.4
- Estimated · #75/184
- Qwen3.5 397B
- 53.1
- Estimated · #68/184
- Basis
- BenchAlign v5.8 lane · 8 vs 6 public rows
- Reading
- Directional only
Instruction following
Directional only- Kimi K2.5
- 85.8
- #48/148
- Qwen3.5 397B
- 79.9
- #59/148
- Basis
- Provisional lane · 0 vs 1 weighted rows
- Reading
- Directional only
Coding
Not comparable- Kimi K2.5
- 38.5
- Estimated · #68/150
- Qwen3.5 397B
- Not ranked
- Basis
- BenchAlign v5.8 lane · 10 vs 6 public rows
- Reading
- Not comparable
Reasoning
Not comparable- Kimi K2.5
- 72.2
- Unranked · 3 rankable rows
- Qwen3.5 397B
- 74.9
- Unranked · 2 rankable rows
- Basis
- Provisional lane · 1 vs 1 weighted rows
- Reading
- Not comparable
Math
Not comparable- Kimi K2.5
- 83.3
- Unranked · 7 rankable rows
- Qwen3.5 397B
- 81.7
- Unranked · 5 rankable rows
- Basis
- Provisional lane · 2 vs 2 weighted rows
- Reading
- Not comparable
| Category | Kimi K2.5 | Qwen3.5 397B | Basis | Reading |
|---|---|---|---|---|
| Multimodal | 70.4#32/54 | 73.6#29/54 | Like-for-likeProvisional lane · 2 vs 2 weighted rows | Qwen3.5 397B leads |
| Multilingual | 87.3#8/17 | 93.8#5/17 | Like-for-likeProvisional lane · 1 vs 1 weighted rows | Qwen3.5 397B leads |
| Agentic | 40.7Estimated · #66/128 | 43.8Estimated · #61/128 | Directional onlyBenchAlign v5.8 lane · 16 vs 14 public rows | Directional only |
| Knowledge | 51.4Estimated · #75/184 | 53.1Estimated · #68/184 | Directional onlyBenchAlign v5.8 lane · 8 vs 6 public rows | Directional only |
| Instruction following | 85.8#48/148 | 79.9#59/148 | Directional onlyProvisional lane · 0 vs 1 weighted rows | Directional only |
| Coding | 38.5Estimated · #68/150 | Not ranked | Not comparableBenchAlign v5.8 lane · 10 vs 6 public rows | Not comparable |
| Reasoning | 72.2Unranked · 3 rankable rows | 74.9Unranked · 2 rankable rows | Not comparableProvisional lane · 1 vs 1 weighted rows | Not comparable |
| Math | 83.3Unranked · 7 rankable rows | 81.7Unranked · 5 rankable rows | Not comparableProvisional lane · 2 vs 2 weighted rows | Not comparable |
Ranks count the models scored in each category’s lane, so the agentic, coding, and knowledge denominators (BenchAlign v5.8) differ from the provisional-lane categories. Unranked scores sit on the provisional lane’s scale but never name a winner. Like-for-like rows within 0.5 points read as a practical tie.
Bars run 0–100Methodology
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 K2.5
- $0.0021
- Fits in one request
- Qwen3.5 397B
- $0.0024
- 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
- Kimi K2.5
- $0.039
- Fits in one request
- Qwen3.5 397B
- $0.0408
- 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
- Kimi K2.5
- $0.162
- Fits in one request
- Cached input priced at the published list-input rate
- Qwen3.5 397B
- $0.168
- Does not fit in one request
- Cached input priced at the published list-input rate
Qwen3.5 397B does not fit this workload in one request. Kimi K2.5 has no published cached-input rate, so cached tokens use its listed input rate. Qwen3.5 397B has no published cached-input rate, so cached tokens use its listed input rate.
Cached input falls back to the list input rate only where a cached rate is unpublished
Specification differences
Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.
SpecificationKimi K2.5Qwen3.5 397B
Context window
Maximum documented context; output-token limits may be lower.
Kimi K2.5
256K
Qwen3.5 397B
128K
API model ID
Kimi K2.5
Not sourced
Qwen3.5 397B
Not sourced
Cached-input rate
A missing cached-input rate falls back to the listed input rate only in the stated workload estimate.
Kimi K2.5
Not published
Qwen3.5 397B
Not published
Documented inputs
Kimi K2.5
Not sourced
Qwen3.5 397B
Not sourced
Documented outputs
Kimi K2.5
Not sourced
Qwen3.5 397B
Not sourced
Provider availability
Kimi K2.5
Not sourced
Qwen3.5 397B
Not sourced
Reasoning profile
Kimi K2.5
Non-Reasoning
Qwen3.5 397B
Non-Reasoning
Weight access
Kimi K2.5
Open Weight
Qwen3.5 397B
Open Weight
License
Kimi K2.5
Open Weight
Qwen3.5 397B
Open Weight
Release date
Kimi K2.5
2026-01-27
Qwen3.5 397B
2026-02-16
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
- Qwen3.5 397B has the higher public point estimate, 55.34 versus 53.05. Their conditional score ranges overlap. These ranges do not establish rank confidence.
- Workload cost
- Repository review: $0.039 vs $0.0408. Cache-heavy agent loop: $0.162 vs $0.168.
- Context tradeoff
- Kimi K2.5 has the larger documented window (256K).
Run the same representative tasks against both endpoints before changing production traffic.
Questions
Which is better, Kimi K2.5 or Qwen3.5 397B?
Qwen3.5 397B has the higher public point estimate, 55.34 versus 53.05. Their conditional score ranges overlap. These ranges do not establish rank confidence. The page therefore keeps the decision tied to the specific documented workload.
Which is better for coding, Kimi K2.5 or Qwen3.5 397B?
Qwen3.5 397B is not ranked on the public lane for coding, so no winner is named for coding.
Which is better for agentic tasks, Kimi K2.5 or Qwen3.5 397B?
Qwen3.5 397B scores higher for agentic tasks on the public lane, 43.8 to 40.7. Kimi K2.5 and Qwen3.5 397B are scored on Estimated evidence for agentic tasks, so the reading is directional rather than like-for-like. BenchLM does not name a winner for agentic tasks from a directional reading; read the shared benchmark rows directly and test the models on the same task set.
Which costs less, Kimi K2.5 or Qwen3.5 397B?
For the stated presets, chat costs $0.0021 on Kimi K2.5 and $0.0024 on Qwen3.5 397B; repository review costs $0.039 and $0.0408; the cache-heavy agent loop costs $0.162 and $0.168. Qwen3.5 397B does not fit this workload in one request. Kimi K2.5 has no published cached-input rate, so cached tokens use its listed input rate. Qwen3.5 397B has no published cached-input rate, so cached tokens use its listed input rate.
Which has the larger context window, Kimi K2.5 or Qwen3.5 397B?
Kimi K2.5 has the larger documented context window: 256K, compared with 128K.
Self-host vs API cost
Estimates at 50,000 req/day · 1000 tokens/req average.
Kimi K2.5
API / mo$2,700
Self-host / mo$7,358
Break-even186M/day
Qwen3.5 397B
API / mo$3,150
Self-host / moNot listed
Break-even—
Proprietary model — self-hosting not applicable.
Benchmark evidence
The full public result ledger is available for audit without forcing a wide desktop table onto a phone.
Browse raw public benchmark evidence75 rows
Agentic
Terminal-Bench 2.0
Kimi K2.550.8%SourceQwen3.5 397B52.5%SourceQwen3.5 397B leads this result
BrowseComp
Kimi K2.560.6%SourceQwen3.5 397B69%SourceQwen3.5 397B leads this result
OSWorld-Verified
Kimi K2.563.3%SourceQwen3.5 397B62.2%SourceKimi K2.5 leads this result
Claw-Eval
Shared sourceKimi K2.552.3%Qwen3.5 397B56.8%Qwen3.5 397B leads this result
QwenClawBench
Shared sourceKimi K2.554.3%Qwen3.5 397B51.8%Kimi K2.5 leads this result
τ³-bench results
Shared sourceKimi K2.565.7%Qwen3.5 397B68.4%Qwen3.5 397B leads this result
DeepSearchQA
Kimi K2.577.1%SourceQwen3.5 397B—Not directly comparable
DeepPlanning
Shared sourceKimi K2.514.4%Qwen3.5 397B37.6%Qwen3.5 397B leads this result
Toolathlon
Kimi K2.527.8%SourceQwen3.5 397B36.3%SourceQwen3.5 397B leads this result
MCP Atlas
Kimi K2.529.5%SourceQwen3.5 397B46.1%SourceQwen3.5 397B leads this result
MCP-Tasks
Shared sourceKimi K2.559.1%Qwen3.5 397B74.2%Qwen3.5 397B leads this result
WideResearch
Kimi K2.572.7%SourceQwen3.5 397B74.0%SourceQwen3.5 397B leads this result
BFCL v4
Kimi K2.568.3%SourceQwen3.5 397B—Not directly comparable
Gert Labs
Shared sourceKimi K2.545.88%Qwen3.5 397B46.76%Qwen3.5 397B leads this result
ResearchClawBench
Shared sourceKimi K2.514.0%Qwen3.5 397B14.2%Qwen3.5 397B leads this result
JobBench
Kimi K2.58.7%SourceQwen3.5 397B—Not directly comparable
VITA-Bench
Kimi K2.5—Qwen3.5 397B43.7%SourceNot directly comparable
Coding
SWE-bench Verified
Kimi K2.576.8%SourceQwen3.5 397B76.2%SourceKimi K2.5 leads this result
SWE-bench Verified*
Kimi K2.570.8%SourceQwen3.5 397B—Not directly comparable
LiveCodeBench v6
Kimi K2.585.0%SourceQwen3.5 397B83.6%SourceKimi K2.5 leads this result
SWE-bench Pro
Kimi K2.550.7%SourceQwen3.5 397B50.9%SourceQwen3.5 397B leads this result
SWE Multilingual
Kimi K2.573%SourceQwen3.5 397B69.3%SourceKimi K2.5 leads this result
SWE-Rebench
Kimi K2.558.5%SourceQwen3.5 397B—Not directly comparable
React Native Evals
Kimi K2.577.2%SourceQwen3.5 397B—Not directly comparable
NL2Repo
Kimi K2.532%SourceQwen3.5 397B32.2%SourceQwen3.5 397B leads this result
SciCode
Kimi K2.548.7%SourceQwen3.5 397B—Not directly comparable
Terminal-Bench 2.0
Kimi K2.550.8%SourceQwen3.5 397B52.5%SourceQwen3.5 397B leads this result
Reasoning
Multimodal
MMMU
Shared sourceKimi K2.584.3%Qwen3.5 397B85.0%Qwen3.5 397B leads this result
MMMU-Pro
Kimi K2.578.5%SourceQwen3.5 397B79%SourceQwen3.5 397B leads this result
RealWorldQA
Shared sourceKimi K2.581.0%Qwen3.5 397B83.9%Qwen3.5 397B leads this result
OmniDocBench 1.5
Kimi K2.588.8%SourceQwen3.5 397B90.8%SourceQwen3.5 397B leads this result
Video-MME
Kimi K2.587.4%SourceQwen3.5 397B—Not directly comparable
Video-MME (with subtitle)
Shared sourceKimi K2.587.4%Qwen3.5 397B87.5%Qwen3.5 397B leads this result
Video-MME (w/o subtitle)
Shared sourceKimi K2.583.2%Qwen3.5 397B83.7%Qwen3.5 397B leads this result
MathVision
Kimi K2.584.2%SourceQwen3.5 397B88.6%SourceQwen3.5 397B leads this result
We-Math
Shared sourceKimi K2.584.7%Qwen3.5 397B87.9%Qwen3.5 397B leads this result
DynaMath
Shared sourceKimi K2.584.4%Qwen3.5 397B86.3%Qwen3.5 397B leads this result
SimpleVQA
Kimi K2.571.2%SourceQwen3.5 397B67.1%SourceKimi K2.5 leads this result
MMLongBench-Doc
Shared sourceKimi K2.558.5%Qwen3.5 397B61.5%Qwen3.5 397B leads this result
CC-OCR
Shared sourceKimi K2.579.7%Qwen3.5 397B82.0%Qwen3.5 397B leads this result
AI2D_TEST
Shared sourceKimi K2.590.8%Qwen3.5 397B93.9%Qwen3.5 397B leads this result
CountBench
Shared sourceKimi K2.594.1%Qwen3.5 397B97.2%Qwen3.5 397B leads this result
RefCOCO (avg)
Shared sourceKimi K2.587.8%Qwen3.5 397B92.3%Qwen3.5 397B leads this result
MMVU
Kimi K2.580.4%SourceQwen3.5 397B—Not directly comparable
VideoMMMU
Shared sourceKimi K2.586.6%Qwen3.5 397B84.7%Kimi K2.5 leads this result
MLVU (M-Avg)
Shared sourceKimi K2.585.0%Qwen3.5 397B86.7%Qwen3.5 397B leads this result
CharXiv
Kimi K2.577.5%SourceQwen3.5 397B80.8%SourceQwen3.5 397B leads this result
ODINW13
Kimi K2.5—Qwen3.5 397B47.0%SourceNot directly comparable
ERQA
Kimi K2.5—Qwen3.5 397B67.5%SourceNot directly comparable
ScreenSpot Pro
Kimi K2.5—Qwen3.5 397B65.6%SourceNot directly comparable
V*
Kimi K2.5—Qwen3.5 397B95.8%SourceNot directly comparable
Knowledge
GPQA
Kimi K2.587.6%SourceQwen3.5 397B88.4%SourceQwen3.5 397B leads this result
GPQA-D
Kimi K2.587.6%SourceQwen3.5 397B—Not directly comparable
SuperGPQA
Shared sourceKimi K2.569.2%Qwen3.5 397B70.4%Qwen3.5 397B leads this result
MMLU-Pro
Kimi K2.587.1%SourceQwen3.5 397B87.8%SourceQwen3.5 397B leads this result
MMLU-Pro (Arcee)
Kimi K2.587.1%SourceQwen3.5 397B—Not directly comparable
MMLU-Redux
Kimi K2.594.5%SourceQwen3.5 397B94.9%SourceQwen3.5 397B leads this result
C-Eval
Kimi K2.594%SourceQwen3.5 397B93%SourceKimi K2.5 leads this result
HLE
Kimi K2.530.1%SourceQwen3.5 397B28.7%SourceKimi K2.5 leads this result
Multilingual
MMLU-ProX
Shared sourceKimi K2.582.3%Qwen3.5 397B84.7%Qwen3.5 397B leads this result
NOVA-63
Shared sourceKimi K2.556.0%Qwen3.5 397B59.1%Qwen3.5 397B leads this result
INCLUDE
Shared sourceKimi K2.583.3%Qwen3.5 397B85.6%Qwen3.5 397B leads this result
PolyMath
Shared sourceKimi K2.543.1%Qwen3.5 397B73.3%Qwen3.5 397B leads this result
MAXIFE
Shared sourceKimi K2.572.8%Qwen3.5 397B88.2%Qwen3.5 397B leads this result
Instruction following
IFEval
Shared sourceKimi K2.593.9%Qwen3.5 397B92.6%Kimi K2.5 leads this result
IFBench
Kimi K2.5—Qwen3.5 397B76.5%SourceNot directly comparable
Math
AIME 2025
Kimi K2.596.1%SourceQwen3.5 397B—Not directly comparable
AIME26
Kimi K2.595.8%SourceQwen3.5 397B93.3%SourceKimi K2.5 leads this result
AIME25 (Arcee)
Kimi K2.596.3%SourceQwen3.5 397B—Not directly comparable
HMMT Feb 2025
Kimi K2.595.4%SourceQwen3.5 397B94.8%SourceKimi K2.5 leads this result
HMMT Nov 2025
Kimi K2.591.1%SourceQwen3.5 397B92.7%SourceQwen3.5 397B leads this result
HMMT Feb 2026
Kimi K2.587.1%SourceQwen3.5 397B87.9%SourceQwen3.5 397B leads this result
MMAnswerBench
Kimi K2.581.8%SourceQwen3.5 397B80.9%SourceKimi K2.5 leads this result
75 public results · 55 shared
Watch Kimi K2.5 vs Qwen3.5 397B
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 5,600+ readers.
Last updated October 17, 2026