Skip to main content

Model comparison

Kimi K2.5 vs Qwen3.6-35B-A3B

Data verified

Head-to-head evidence from 40 shared benchmark results across 7 categories. Overall scores shown here use the public BenchAlign v5 ranking lane.

Moonshot AI
59.58/100
Margin
8.2pts
← winning
51.36/100
3 category wins2 category wins

Verified leaderboard positions: Kimi K2.5 #22; Qwen3.6-35B-A3B #31

BenchAlign evidence: Kimi K2.5 supported; Qwen3.6-35B-A3B estimated. Intervals and evidence labels describe ranking uncertainty, not a guarantee for a specific workload.

Evidence parity. Kimi K2.5 and Qwen3.6-35B-A3B share 40 comparable benchmark results. 5 of 8 categories are comparable. 24 results are unique to Kimi K2.5; 18 to Qwen3.6-35B-A3B.

Updated July 15, 2026
Shared results
40
Kimi K2.5 only
24
Qwen3.6-35B-A3B only
18
Comparable categories
5 / 8

Pick Kimi K2.5 if you want the stronger benchmark profile. Qwen3.6-35B-A3B only becomes the better choice if mathematics is the priority or you need the larger 262K context window.

Confidence note. This is a partial-evidence comparison with 40 shared benchmark results across 7 evidence categories; 5 of 8 categories currently have scoreable aggregates for both models. Treat the verdict as directional until coverage is more balanced.

Why this result

Kimi K2.5 has the cleaner provisional overall profile here, landing at 61 versus 59. It is a real lead, but still close enough that category-level strengths matter more than the headline number.

Kimi K2.5's sharpest advantage is in knowledge, where it averages 57.2 against 51.8. The single biggest benchmark swing on the page is HLE, 30.1% to 21.4%. Qwen3.6-35B-A3B does hit back in mathematics, so the answer changes if that is the part of the workload you care about most.

Qwen3.6-35B-A3B is the reasoning model in the pair, while Kimi K2.5 is not. That usually helps on harder chain-of-thought-heavy tests, but it can also mean more latency and more token spend in real use. Qwen3.6-35B-A3B gives you the larger context window at 262K, compared with 256K for Kimi K2.5.

Category breakdown

Exact category averages are shown below. Not measured means BenchLM does not have enough sourced public coverage for that model and category.

Category scores and score margins for Kimi K2.5 and Qwen3.6-35B-A3B
CategoryKimi K2.5ΔQwen3.6-35B-A3B
MathKimi K2.560.6Margin 27.6Qwen3.6-35B-A3B88.2
CodingKimi K2.559.4Margin 14.4Qwen3.6-35B-A3B73.8
KnowledgeKimi K2.557.2Margin 5.4Qwen3.6-35B-A3B51.8
AgenticKimi K2.555.0Margin 3.5Qwen3.6-35B-A3B51.5
MultimodalKimi K2.578.5Margin 2.2Qwen3.6-35B-A3B76.3
ReasoningKimi K2.561.0MarginNo overlapQwen3.6-35B-A3BNot measured
MultilingualKimi K2.582.3MarginNo overlapQwen3.6-35B-A3BNot measured
Inst. FollowingKimi K2.593.9MarginNo overlapQwen3.6-35B-A3BNot measured

Decisive benchmark drivers

The largest measured benchmark gaps in this matchup, with exact reported values.

More
A · Kimi K2.5B · Qwen3.6-35B-A3B
  1. HLE

    Knowledge
    Source ↗
    A 30.1%B 21.4%
    Winner: Kimi K2.5Δ 8.7
    HLE: Kimi K2.5 scored 30.1%; Qwen3.6-35B-A3B scored 21.4%. Kimi K2.5 wins this benchmark.
  2. SuperGPQA

    Knowledge
    Source ↗
    A 69.2%B 64.7%
    Winner: Kimi K2.5Δ 4.5
    SuperGPQA: Kimi K2.5 scored 69.2%; Qwen3.6-35B-A3B scored 64.7%. Kimi K2.5 wins this benchmark.
  3. HMMT Feb 2026

    Math
    Source ↗
    A 87.1%B 83.6%
    Winner: Kimi K2.5Δ 3.5
    HMMT Feb 2026: Kimi K2.5 scored 87.1%; Qwen3.6-35B-A3B scored 83.6%. Kimi K2.5 wins this benchmark.
  4. SWE-bench Verified

    Coding
    Source ↗
    A 76.8%B 73.4%
    Winner: Kimi K2.5Δ 3.4
    SWE-bench Verified: Kimi K2.5 scored 76.8%; Qwen3.6-35B-A3B scored 73.4%. Kimi K2.5 wins this benchmark.
  5. MMMU-Pro

    Multimodal
    Source ↗
    A 78.5%B 75.3%
    Winner: Kimi K2.5Δ 3.2
    MMMU-Pro: Kimi K2.5 scored 78.5%; Qwen3.6-35B-A3B scored 75.3%. Kimi K2.5 wins this benchmark.

Operational comparison

Runtime and commercial metrics are compared only when both models have a complete sourced value.

MetricKimi K2.5Qwen3.6-35B-A3BComparison
Input / output priceUSD per 1M tokensKimi K2.5$0.6 input / $3 outputQwen3.6-35B-A3BNot availableA complete price comparison is not available.
Generation speedtokens per secondKimi K2.545 tok/sQwen3.6-35B-A3BNot availableA complete speed comparison is not available.
First-answer latencyseconds to first tokenKimi K2.52.38 sQwen3.6-35B-A3BNot availableA complete latency comparison is not available.
Context windowmaximum listed tokensKimi K2.5256KQwen3.6-35B-A3B262KQwen3.6-35B-A3B lists the larger context window.

Benchmark Deep Dive

AgenticKimi K2.5 wins
BenchmarkKimi K2.5Qwen3.6-35B-A3BResult
Terminal-Bench 2.0Source 50.8%51.5%Qwen3.6-35B-A3B leads
BrowseCompSource 60.6%Not comparable
Claw-EvalSource 52.3%68.7%Qwen3.6-35B-A3B leads
QwenClawBenchSource 54.3%52.6%Kimi K2.5 leads
τ³-bench resultsSource 65.7%67.2%Qwen3.6-35B-A3B leads
DeepSearchQASource 77.1%Not comparable
DeepPlanningSource 14.4%25.9%Qwen3.6-35B-A3B leads
ToolathlonSource 27.8%26.9%Kimi K2.5 leads
MCP AtlasSource 29.5%62.8%Qwen3.6-35B-A3B leads
MCP-TasksSource 59.1%Not comparable
WideResearchSource 72.7%60.1%Kimi K2.5 leads
τ²-bench resultsSource 95.9%95.3%Kimi K2.5 leads
APEX-Agents-AASource 11.5%Not comparable
Gert LabsSource 45.88%42.65%Kimi K2.5 leads
ResearchClawBenchSource 14.0%Not comparable
JobBenchSource 8.7%Not comparable
AA Agentic IndexSource 21.7%21.4%Kimi K2.5 leads
GDPval-AASource 25.4%27.4%Qwen3.6-35B-A3B leads
GDPval-AASource 10091049Qwen3.6-35B-A3B leads
QwenWebBenchSource 1397Not comparable
VITA-BenchSource 35.6%Not comparable
CodingQwen3.6-35B-A3B wins
BenchmarkKimi K2.5Qwen3.6-35B-A3BResult
SWE-bench VerifiedSource 76.8%73.4%Kimi K2.5 leads
SWE-bench Verified*Source 70.8%Not comparable
LiveCodeBench v6Source 85.0%Not comparable
SWE-bench ProSource 50.7%49.5%Kimi K2.5 leads
SWE MultilingualSource 73%67.2%Kimi K2.5 leads
SWE-RebenchSource 58.5%Not comparable
React Native EvalsSource 77.2%Not comparable
SciCodeSource 48.7%Not comparable
Terminal-Bench HardSource 34.8%34.8%Tie
AA-SciCodeSource 49.0%35.8%Kimi K2.5 leads
AA Coding IndexSource 46.8%41.9%Kimi K2.5 leads
Terminal-Bench 2.0Source 51.5%Not comparable
LiveCodeBenchSource 80.4%Not comparable
NL2RepoSource 29.4%Not comparable
Reasoning
BenchmarkKimi K2.5Qwen3.6-35B-A3BResult
LongBench v2Source 61%Not comparable
AA-LCRSource 65.3%63.7%Kimi K2.5 leads
CritPtSource 3.1%0.3%Kimi K2.5 leads
KnowledgeKimi K2.5 wins
BenchmarkKimi K2.5Qwen3.6-35B-A3BResult
GPQASource 87.6%86%Kimi K2.5 leads
GPQA-DSource 87.6%Not comparable
SuperGPQASource 69.2%64.7%Kimi K2.5 leads
MMLU-ProSource 87.1%85.2%Kimi K2.5 leads
MMLU-Pro (Arcee)Source 87.1%Not comparable
HLESource 30.1%21.4%Kimi K2.5 leads
Artificial Analysis Intelligence IndexSource 35.4%31.6%Kimi K2.5 leads
AA-GPQA DiamondSource 87.9%84.1%Kimi K2.5 leads
AA-HLESource 29.4%20.2%Kimi K2.5 leads
AA-Omniscience IndexSource -8.1%-21.4%Kimi K2.5 leads
AA-Omniscience AccuracySource 34.3%18.9%Kimi K2.5 leads
AA-Omniscience Hallucination RateSource 64.6%49.7%Qwen3.6-35B-A3B leads
C-EvalSource 90%Not comparable
MathQwen3.6-35B-A3B wins
BenchmarkKimi K2.5Qwen3.6-35B-A3BResult
AIME 2025Source 96.1%Not comparable
AIME26Source 95.8%92.7%Kimi K2.5 leads
AIME25 (Arcee)Source 96.3%Not comparable
HMMT Feb 2025Source 95.4%90.7%Kimi K2.5 leads
HMMT Nov 2025Source 91.1%89.1%Kimi K2.5 leads
HMMT Feb 2026Source 87.1%83.6%Kimi K2.5 leads
MMAnswerBenchSource 81.8%78.9%Kimi K2.5 leads
FrontierMath v2 (Tiers 1-3)Source 27.900%Not comparable
FrontierMath v2 (Tier 4)Source 4.200%Not comparable
Multilingual
BenchmarkKimi K2.5Qwen3.6-35B-A3BResult
MMLU-ProXSource 82.3%Not comparable
NOVA-63Source 56.0%Not comparable
MultimodalKimi K2.5 wins
BenchmarkKimi K2.5Qwen3.6-35B-A3BResult
MMMU-ProSource 78.5%75.3%Kimi K2.5 leads
Video-MMESource 87.4%Not comparable
MMVUSource 80.4%Not comparable
VideoMMMUSource 86.6%83.7%Kimi K2.5 leads
AA-MMMU-ProSource 75.4%75.0%Kimi K2.5 leads
Design Arena WebsiteSource 1284Not comparable
MMMUSource 81.7%Not comparable
RealWorldQASource 85.3%Not comparable
OmniDocBench 1.5Source 89.9%Not comparable
CharXivSource 78%Not comparable
SimpleVQASource 58.9%Not comparable
CC-OCRSource 81.9%Not comparable
AI2D_TESTSource 92.7%Not comparable
RefCOCO (avg)Source 92.0%Not comparable
ODINW13Source 50.8%Not comparable
Video-MME (with subtitle)Source 86.6%Not comparable
Video-MME (w/o subtitle)Source 82.5%Not comparable
MLVU (M-Avg)Source 86.2%Not comparable
Inst. Following
BenchmarkKimi K2.5Qwen3.6-35B-A3BResult
IFEvalSource 93.9%Not comparable
AA-IFBenchSource 70.2%64.4%Kimi K2.5 leads
Frequently Asked Questions (6)

Which is better, Kimi K2.5 or Qwen3.6-35B-A3B?

Kimi K2.5 is ahead on BenchLM's provisional leaderboard, 61 to 59. The biggest single separator in this matchup is HLE, where the scores are 30.1% and 21.4%.

Which is better for knowledge tasks, Kimi K2.5 or Qwen3.6-35B-A3B?

Kimi K2.5 has the edge for knowledge tasks in this comparison, averaging 57.2 versus 51.8. Inside this category, AA-Omniscience Accuracy is the benchmark that creates the most daylight between them.

Which is better for coding, Kimi K2.5 or Qwen3.6-35B-A3B?

Qwen3.6-35B-A3B has the edge for coding in this comparison, averaging 73.8 versus 59.4. Inside this category, AA-SciCode is the benchmark that creates the most daylight between them.

Which is better for math, Kimi K2.5 or Qwen3.6-35B-A3B?

Qwen3.6-35B-A3B has the edge for math in this comparison, averaging 88.2 versus 60.6. Inside this category, HMMT Feb 2025 is the benchmark that creates the most daylight between them.

Which is better for agentic tasks, Kimi K2.5 or Qwen3.6-35B-A3B?

Kimi K2.5 has the edge for agentic tasks in this comparison, averaging 55 versus 51.5. Inside this category, GDPval-AA is the benchmark that creates the most daylight between them.

Which is better for multimodal and grounded tasks, Kimi K2.5 or Qwen3.6-35B-A3B?

Kimi K2.5 has the edge for multimodal and grounded tasks in this comparison, averaging 78.5 versus 76.3. Inside this category, MMMU-Pro is the benchmark that creates the most daylight between them.

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$5,221
Break-even132M/day
Qwen3.6-35B-A3B
API / mo$0
Self-host / moNot listed
Break-even
Proprietary model — self-hosting not applicable.
Model the full break-even

Related Comparisons

Last updated: July 15, 2026

The AI models change fast. We track them for you.

A weekly brief for engineers and researchers covering new models, ranking shifts, and pricing changes.

Free. No spam. Unsubscribe anytime.