Skip to main content

Model comparison

Kimi K2.5 vs Step 3.7 Flash

Data verified

Head-to-head evidence from 26 shared benchmark results across 6 categories. Overall scores shown here use the public BenchAlign v5 ranking lane.

Moonshot AI
59.58/100
Margin
8.8pts
← winning
50.76/100
1 category wins1 category wins

Verified leaderboard positions: Kimi K2.5 #22; Step 3.7 Flash unranked

BenchAlign evidence: Kimi K2.5 supported; Step 3.7 Flash estimated. Intervals and evidence labels describe ranking uncertainty, not a guarantee for a specific workload.

Evidence parity. Kimi K2.5 and Step 3.7 Flash share 26 comparable benchmark results. 2 of 8 categories are comparable. 38 results are unique to Kimi K2.5; 4 to Step 3.7 Flash.

Updated July 16, 2026
Shared results
26
Kimi K2.5 only
38
Step 3.7 Flash only
4
Comparable categories
2 / 8

Pick Kimi K2.5 if you want the stronger benchmark profile. Step 3.7 Flash only becomes the better choice if agentic is the priority or you want the cheaper token bill.

Confidence note. This is a partial-evidence comparison with 26 shared benchmark results across 6 evidence categories; 2 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 is clearly ahead on the provisional aggregate, 61 to 57. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.

Kimi K2.5's sharpest advantage is in coding, where it averages 59.4 against 56.3. The single biggest benchmark swing on the page is BrowseComp, 60.6% to 75.8%. Step 3.7 Flash does hit back in agentic, so the answer changes if that is the part of the workload you care about most.

Kimi K2.5 is also the more expensive model on tokens at $0.60 input / $3.00 output per 1M tokens, versus $0.20 input / $1.15 output per 1M tokens for Step 3.7 Flash. That is roughly 2.6x on output cost alone. Step 3.7 Flash 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.

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 Step 3.7 Flash
CategoryKimi K2.5ΔStep 3.7 Flash
AgenticKimi K2.555.0Margin 11.4Step 3.7 Flash66.4
CodingKimi K2.559.4Margin 3.1Step 3.7 Flash56.3
ReasoningKimi K2.561.0MarginNo overlapStep 3.7 FlashNot measured
KnowledgeKimi K2.557.2MarginNo overlapStep 3.7 FlashNot measured
MathKimi K2.560.6MarginNo overlapStep 3.7 FlashNot measured
MultilingualKimi K2.582.3MarginNo overlapStep 3.7 FlashNot measured
MultimodalKimi K2.578.5MarginNo overlapStep 3.7 FlashNot measured
Inst. FollowingKimi K2.593.9MarginNo overlapStep 3.7 FlashNot measured

Decisive benchmark drivers

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

More
A · Kimi K2.5B · Step 3.7 Flash
  1. BrowseComp

    Agentic
    Source ↗
    A 60.6%B 75.8%
    Winner: Step 3.7 FlashΔ 15.2
    BrowseComp: Kimi K2.5 scored 60.6%; Step 3.7 Flash scored 75.8%. Step 3.7 Flash wins this benchmark.
  2. Terminal-Bench 2.0

    Agentic
    Source ↗
    A 50.8%B 59.5%
    Winner: Step 3.7 FlashΔ 8.7
    Terminal-Bench 2.0: Kimi K2.5 scored 50.8%; Step 3.7 Flash scored 59.5%. Step 3.7 Flash wins this benchmark.
  3. SWE-bench Pro

    Coding
    Source ↗
    A 50.7%B 56.3%
    Winner: Step 3.7 FlashΔ 5.6
    SWE-bench Pro: Kimi K2.5 scored 50.7%; Step 3.7 Flash scored 56.3%. Step 3.7 Flash wins this benchmark.

Operational comparison

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

MetricKimi K2.5Step 3.7 FlashComparison
Input / output priceUSD per 1M tokensKimi K2.5$0.6 input / $3 outputStep 3.7 Flash$0.2 input / $1.15 outputStep 3.7 Flash has the lower combined listed price.
Generation speedtokens per secondKimi K2.545 tok/sStep 3.7 FlashNot availableA complete speed comparison is not available.
First-answer latencyseconds to first tokenKimi K2.52.38 sStep 3.7 FlashNot availableA complete latency comparison is not available.
Context windowmaximum listed tokensKimi K2.5256KStep 3.7 Flash256KListed context windows are equal.

Benchmark Deep Dive

AgenticStep 3.7 Flash wins
BenchmarkKimi K2.5Step 3.7 FlashResult
Terminal-Bench 2.0Source 50.8%59.5%Step 3.7 Flash leads
BrowseCompSource 60.6%75.8%Step 3.7 Flash leads
Claw-EvalSource 52.3%67.1%Step 3.7 Flash leads
QwenClawBenchSource 54.3%Not comparable
τ³-bench resultsSource 65.7%Not comparable
DeepSearchQASource 77.1%92.8%Step 3.7 Flash leads
DeepPlanningSource 14.4%Not comparable
ToolathlonSource 27.8%49.5%Step 3.7 Flash leads
MCP AtlasSource 29.5%Not comparable
MCP-TasksSource 59.1%Not comparable
WideResearchSource 72.7%Not comparable
τ²-bench resultsSource 95.9%98.5%Step 3.7 Flash leads
APEX-Agents-AASource 11.5%14.8%Step 3.7 Flash leads
Gert LabsSource 45.88%51.57%Step 3.7 Flash leads
ResearchClawBenchSource 14.0%Not comparable
JobBenchSource 8.7%Not comparable
AA Agentic IndexSource 21.7%21.5%Kimi K2.5 leads
GDPval-AASource 25.4%25.9%Step 3.7 Flash leads
GDPval-AASource 10091017Step 3.7 Flash leads
HLE w/ toolsSource 47.2%Not comparable
CodingKimi K2.5 wins
BenchmarkKimi K2.5Step 3.7 FlashResult
SWE-bench VerifiedSource 76.8%Not comparable
SWE-bench Verified*Source 70.8%Not comparable
LiveCodeBench v6Source 85.0%Not comparable
SWE-bench ProSource 50.7%56.3%Step 3.7 Flash leads
SWE MultilingualSource 73%Not comparable
SWE-RebenchSource 58.5%Not comparable
React Native EvalsSource 77.2%Not comparable
SciCodeSource 48.7%Not comparable
Terminal-Bench HardSource 34.8%35.6%Step 3.7 Flash leads
AA-SciCodeSource 49.0%40.0%Kimi K2.5 leads
AA Coding IndexSource 46.8%39.6%Kimi K2.5 leads
Terminal-Bench 2.0Source 59.5%Not comparable
Reasoning
BenchmarkKimi K2.5Step 3.7 FlashResult
LongBench v2Source 61%Not comparable
AA-LCRSource 65.3%63.7%Kimi K2.5 leads
CritPtSource 3.1%2.3%Kimi K2.5 leads
Knowledge
BenchmarkKimi K2.5Step 3.7 FlashResult
GPQASource 87.6%Not comparable
GPQA-DSource 87.6%Not comparable
SuperGPQASource 69.2%Not comparable
MMLU-ProSource 87.1%Not comparable
MMLU-Pro (Arcee)Source 87.1%Not comparable
HLESource 30.1%Not comparable
Artificial Analysis Intelligence IndexSource 35.4%30.3%Kimi K2.5 leads
AA-GPQA DiamondSource 87.9%80.9%Kimi K2.5 leads
AA-HLESource 29.4%19.9%Kimi K2.5 leads
AA-Omniscience IndexSource -8.1%-37.5%Kimi K2.5 leads
AA-Omniscience AccuracySource 34.3%25.4%Kimi K2.5 leads
AA-Omniscience Hallucination RateSource 64.6%84.4%Kimi K2.5 leads
Math
BenchmarkKimi K2.5Step 3.7 FlashResult
AIME 2025Source 96.1%Not comparable
AIME26Source 95.8%Not comparable
AIME25 (Arcee)Source 96.3%Not comparable
HMMT Feb 2025Source 95.4%Not comparable
HMMT Nov 2025Source 91.1%Not comparable
HMMT Feb 2026Source 87.1%Not comparable
MMAnswerBenchSource 81.8%Not comparable
FrontierMath v2 (Tiers 1-3)Source 27.900%Not comparable
FrontierMath v2 (Tier 4)Source 4.200%Not comparable
Multilingual
BenchmarkKimi K2.5Step 3.7 FlashResult
MMLU-ProXSource 82.3%Not comparable
NOVA-63Source 56.0%Not comparable
Multimodal
BenchmarkKimi K2.5Step 3.7 FlashResult
MMMU-ProSource 78.5%Not comparable
Video-MMESource 87.4%Not comparable
MMVUSource 80.4%Not comparable
VideoMMMUSource 86.6%Not comparable
AA-MMMU-ProSource 75.4%75.3%Kimi K2.5 leads
Design Arena WebsiteSource 12841218Kimi K2.5 leads
SimpleVQASource 79.2%Not comparable
V*Source 95.3%Not comparable
Inst. Following
BenchmarkKimi K2.5Step 3.7 FlashResult
IFEvalSource 93.9%Not comparable
AA-IFBenchSource 70.2%67.3%Kimi K2.5 leads
Frequently Asked Questions (3)

Which is better, Kimi K2.5 or Step 3.7 Flash?

Kimi K2.5 is ahead on BenchLM's provisional leaderboard, 61 to 57. The biggest single separator in this matchup is BrowseComp, where the scores are 60.6% and 75.8%.

Which is better for coding, Kimi K2.5 or Step 3.7 Flash?

Kimi K2.5 has the edge for coding in this comparison, averaging 59.4 versus 56.3. Inside this category, AA-SciCode is the benchmark that creates the most daylight between them.

Which is better for agentic tasks, Kimi K2.5 or Step 3.7 Flash?

Step 3.7 Flash has the edge for agentic tasks in this comparison, averaging 66.4 versus 55. Inside this category, Toolathlon 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
Step 3.7 Flash
API / mo$1,012
Self-host / moNot listed
Break-even
Proprietary model — self-hosting not applicable.
Model the full break-even

Related Comparisons

Last updated: July 16, 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.