Agentic
Like-for-like- Kimi K2.5
- 55.0
- Step 3.7 Flash
- 66.4
- Weighted basis
- 2 vs 2 rows
- Reading
- Step 3.7 Flash leads
Kimi K2.5 vs Step 3.7 Flash
Updated August 30, 2026. Public scores include evidence status and uncertainty. They are not guarantees for a specific workload.
Decision reading
Kimi K2.5 has the higher public score estimate, 59.14 versus 51.05, but the 90% score intervals overlap. Treat that as a lead, not a settled winner.
7 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.
Agentic work
Tool use, computer use, and multi-step task completion
Step 3.7 Flash
Step 3.7 Flash leads on the same 2 weighted benchmark rows.
Confidence: limited
Chat turn cost
1K fresh input + 500 output tokens
Step 3.7 Flash
Step 3.7 Flash has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates
Cache-heavy agent loop cost
200K cached + 20K fresh input + 10K output tokens
Step 3.7 Flash
Step 3.7 Flash has the lower estimated token cost for this stated workload. Kimi K2.5 has no published cached-input rate, so cached tokens use its listed input rate. Step 3.7 Flash has no published cached-input rate, so cached tokens use its listed input rate.
Confidence: rate-fallback
Show secondary and unsupported calls
Repository review cost
50K fresh input + 3K output tokens
Step 3.7 Flash
Step 3.7 Flash 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
Not enough matched evidence
The category averages use different weighted benchmark sets, so they are directional rather than like-for-like.
Confidence: limited
Long documents
Prompts that approach the documented context limit
No clear pick
The documented context windows are equal.
Confidence: documented
What is actually comparable
Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.
38 Kimi K2.5 only
7 Shared
4 Step 3.7 Flash only
- Shared results
- 7
- Kimi K2.5 only
- 38
- Step 3.7 Flash only
- 4
- Like-for-like categories
- 1 / 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.
Coding
Directional only- Kimi K2.5
- 59.4
- Step 3.7 Flash
- 56.3
- Weighted basis
- 4 vs 1 rows
- Reading
- Directional only
Reasoning
Not comparable- Kimi K2.5
- 61.0
- Step 3.7 Flash
- Not measured
- Weighted basis
- 1 vs 0 rows
- Reading
- Not comparable
Knowledge
Not comparable- Kimi K2.5
- 56.9
- Step 3.7 Flash
- Not measured
- Weighted basis
- 4 vs 0 rows
- Reading
- Not comparable
Math
Not comparable- Kimi K2.5
- 60.6
- Step 3.7 Flash
- Not measured
- Weighted basis
- 4 vs 0 rows
- Reading
- Not comparable
Multilingual
Not comparable- Kimi K2.5
- 82.3
- Step 3.7 Flash
- Not measured
- Weighted basis
- 1 vs 0 rows
- Reading
- Not comparable
Multimodal
Not comparable- Kimi K2.5
- 78.5
- Step 3.7 Flash
- Not measured
- Weighted basis
- 1 vs 0 rows
- Reading
- Not comparable
Instruction following
Not comparable- Kimi K2.5
- 93.9
- Step 3.7 Flash
- Not measured
- Weighted basis
- 1 vs 0 rows
- Reading
- Not comparable
| Category | Kimi K2.5 | Step 3.7 Flash | Weighted basis | Reading |
|---|---|---|---|---|
| Agentic | 55.0 | 66.4 | Like-for-like2 vs 2 rows | Step 3.7 Flash leads |
| Coding | 59.4 | 56.3 | Directional only4 vs 1 rows | Directional only |
| Reasoning | 61.0 | Not measured | Not comparable1 vs 0 rows | Not comparable |
| Knowledge | 56.9 | Not measured | Not comparable4 vs 0 rows | Not comparable |
| Math | 60.6 | Not measured | Not comparable4 vs 0 rows | Not comparable |
| Multilingual | 82.3 | Not measured | Not comparable1 vs 0 rows | Not comparable |
| Multimodal | 78.5 | Not measured | Not comparable1 vs 0 rows | Not comparable |
| Instruction following | 93.9 | Not measured | Not comparable1 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.
BrowseComp
Agentic
Terminal-Bench 2.0
Agentic
SWE-bench Pro
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
- Kimi K2.5
- $0.0021
- Fits in one request
- Step 3.7 Flash
- $0.00077
- Fits in one request
Step 3.7 Flash 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
- Step 3.7 Flash
- $0.01345
- Fits in one request
Step 3.7 Flash 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
- Step 3.7 Flash
- $0.0555
- Fits in one request
- Cached input priced at the published list-input rate
Step 3.7 Flash has the lower modeled cost
Kimi K2.5 has no published cached-input rate, so cached tokens use its listed input rate. Step 3.7 Flash has no published cached-input rate, so cached tokens use its listed input rate.
Specification differences
Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.
SpecificationKimi K2.5Step 3.7 Flash
Context window
Maximum documented context; output-token limits may be lower.
Kimi K2.5
256K
Step 3.7 Flash
256K
API model ID
Kimi K2.5
Not sourced
Step 3.7 Flash
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
Step 3.7 Flash
Not published
Documented inputs
Kimi K2.5
Not sourced
Step 3.7 Flash
Not sourced
Documented outputs
Kimi K2.5
Not sourced
Step 3.7 Flash
Not sourced
Provider availability
Kimi K2.5
Not sourced
Step 3.7 Flash
Not sourced
Reasoning profile
Kimi K2.5
Non-Reasoning
Step 3.7 Flash
Reasoning
Weight access
Kimi K2.5
Open Weight
Step 3.7 Flash
Open Weight
License
Kimi K2.5
Open Weight
Step 3.7 Flash
Open Weight
Release date
Kimi K2.5
2026-02-01
Step 3.7 Flash
2026-05-29
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 K2.5 has the higher public score estimate, 59.14 versus 51.05, but the 90% score intervals overlap.
- Workload cost
- Repository review: $0.039 vs $0.01345. Cache-heavy agent loop: $0.162 vs $0.0555.
- Context tradeoff
- Both models list 256K.
Run the same representative tasks against both endpoints before changing production traffic.
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.
Benchmark evidence
The full public result ledger is available for audit without forcing a wide desktop table onto a phone.
Browse raw public benchmark evidence49 rows
Agentic
Terminal-Bench 2.0
Kimi K2.550.8%SourceStep 3.7 Flash59.5%SourceStep 3.7 Flash leads this result
BrowseComp
Kimi K2.560.6%SourceStep 3.7 Flash75.8%SourceStep 3.7 Flash leads this result
Claw-Eval
Kimi K2.552.3%SourceStep 3.7 Flash67.1%SourceStep 3.7 Flash leads this result
QwenClawBench
Kimi K2.554.3%SourceStep 3.7 Flash—Not directly comparable
τ³-bench results
Kimi K2.565.7%SourceStep 3.7 Flash—Not directly comparable
DeepSearchQA
Kimi K2.577.1%SourceStep 3.7 Flash92.8%SourceStep 3.7 Flash leads this result
DeepPlanning
Kimi K2.514.4%SourceStep 3.7 Flash—Not directly comparable
Toolathlon
Kimi K2.527.8%SourceStep 3.7 Flash49.5%SourceStep 3.7 Flash leads this result
MCP Atlas
Kimi K2.529.5%SourceStep 3.7 Flash—Not directly comparable
MCP-Tasks
Kimi K2.559.1%SourceStep 3.7 Flash—Not directly comparable
WideResearch
Kimi K2.572.7%SourceStep 3.7 Flash—Not directly comparable
Gert Labs
Shared sourceKimi K2.545.88%Step 3.7 Flash51.57%Step 3.7 Flash leads this result
ResearchClawBench
Kimi K2.514.0%SourceStep 3.7 Flash—Not directly comparable
JobBench
Kimi K2.58.7%SourceStep 3.7 Flash—Not directly comparable
HLE w/ tools
Kimi K2.5—Step 3.7 Flash47.2%SourceNot directly comparable
Coding
SWE-bench Verified
Kimi K2.576.8%SourceStep 3.7 Flash—Not directly comparable
SWE-bench Verified*
Kimi K2.570.8%SourceStep 3.7 Flash—Not directly comparable
LiveCodeBench v6
Kimi K2.585.0%SourceStep 3.7 Flash—Not directly comparable
SWE-bench Pro
Kimi K2.550.7%SourceStep 3.7 Flash56.3%SourceStep 3.7 Flash leads this result
SWE Multilingual
Kimi K2.573%SourceStep 3.7 Flash—Not directly comparable
SWE-Rebench
Kimi K2.558.5%SourceStep 3.7 Flash—Not directly comparable
React Native Evals
Kimi K2.577.2%SourceStep 3.7 Flash—Not directly comparable
SciCode
Kimi K2.548.7%SourceStep 3.7 Flash—Not directly comparable
Terminal-Bench 2.0
Kimi K2.5—Step 3.7 Flash59.5%SourceNot directly comparable
Reasoning
LongBench v2
Kimi K2.561%SourceStep 3.7 Flash—Not directly comparable
Knowledge
GPQA
Kimi K2.587.6%SourceStep 3.7 Flash—Not directly comparable
GPQA-D
Kimi K2.587.6%SourceStep 3.7 Flash—Not directly comparable
SuperGPQA
Kimi K2.569.2%SourceStep 3.7 Flash—Not directly comparable
MMLU-Pro
Kimi K2.587.1%SourceStep 3.7 Flash—Not directly comparable
MMLU-Pro (Arcee)
Kimi K2.587.1%SourceStep 3.7 Flash—Not directly comparable
HLE
Kimi K2.530.1%SourceStep 3.7 Flash—Not directly comparable
Math
AIME 2025
Kimi K2.596.1%SourceStep 3.7 Flash—Not directly comparable
AIME26
Kimi K2.595.8%SourceStep 3.7 Flash—Not directly comparable
AIME25 (Arcee)
Kimi K2.596.3%SourceStep 3.7 Flash—Not directly comparable
HMMT Feb 2025
Kimi K2.595.4%SourceStep 3.7 Flash—Not directly comparable
HMMT Nov 2025
Kimi K2.591.1%SourceStep 3.7 Flash—Not directly comparable
HMMT Feb 2026
Kimi K2.587.1%SourceStep 3.7 Flash—Not directly comparable
MMAnswerBench
Kimi K2.581.8%SourceStep 3.7 Flash—Not directly comparable
FrontierMath v2 (Tiers 1-3)
Kimi K2.527.900%SourceStep 3.7 Flash—Not directly comparable
FrontierMath v2 (Tier 4)
Kimi K2.54.200%SourceStep 3.7 Flash—Not directly comparable
Multilingual
Multimodal
MMMU-Pro
Kimi K2.578.5%SourceStep 3.7 Flash—Not directly comparable
Video-MME
Kimi K2.587.4%SourceStep 3.7 Flash—Not directly comparable
MMVU
Kimi K2.580.4%SourceStep 3.7 Flash—Not directly comparable
VideoMMMU
Kimi K2.586.6%SourceStep 3.7 Flash—Not directly comparable
SimpleVQA
Kimi K2.5—Step 3.7 Flash79.2%SourceNot directly comparable
V*
Kimi K2.5—Step 3.7 Flash95.3%SourceNot directly comparable
Instruction following
IFEval
Kimi K2.593.9%SourceStep 3.7 Flash—Not directly comparable
Frequently asked questions
Which is better, Kimi K2.5 or Step 3.7 Flash?
Kimi K2.5 has the higher public score estimate, 59.14 versus 51.05, but the 90% score intervals overlap. The higher estimate is not a decisive winner because the uncertainty ranges overlap.
Which is better for coding, Kimi K2.5 or Step 3.7 Flash?
The current coding averages use different weighted benchmark sets, so BenchLM does not name a winner from them. Read the shared benchmark rows directly and test the models on the same task set.
Which is better for agentic tasks, Kimi K2.5 or Step 3.7 Flash?
Step 3.7 Flash leads the like-for-like agentic tasks comparison across 2 shared weighted benchmark rows.
Which costs less, Kimi K2.5 or Step 3.7 Flash?
For the stated presets, chat costs $0.0021 on Kimi K2.5 and $0.00077 on Step 3.7 Flash; repository review costs $0.039 and $0.01345; the cache-heavy agent loop costs $0.162 and $0.0555. Kimi K2.5 has no published cached-input rate, so cached tokens use its listed input rate. Step 3.7 Flash has no published cached-input rate, so cached tokens use its listed input rate.
Which has the larger context window, Kimi K2.5 or Step 3.7 Flash?
Both models list the same context window, 256K.
Related comparisons
Last updated August 30, 2026
Watch Kimi K2.5 vs Step 3.7 Flash
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 2,000+ readers.