Skip to main content
Radar

Every change to the models you run, with its source and its date. Releases, price changes, retirements, API changes, and incidents.Every change to the models you run, with its source.

Follow model changes
BenchLM

Kimi K3 vs Step 5 Preview

Decision reading

At least one model is not scored in the current public ranking lane, so the page does not name an overall quality winner.

14 results are shared. Category rows resting on Estimated evidence or different benchmark sets are marked directional and do not name a winner.

Moonshot AI logo
Model A
Kimi K3

Moonshot AI

74.3/100

Supported · Public rank #8

90% interval 71.3–77.3

StepFun logo
Model B
Step 5 Preview

StepFun

—

Evidence status unavailable

90% interval unavailable

Updated September 21, 2026. We do not rank this pair: at least one has no public score. Public scores include evidence status and uncertainty.

Share or export

Share on XLinkedInSocial cardCSVJSON

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

  • Chat turn cost

    1K fresh input + 500 output tokens

    Step 5 Preview

    Step 5 Preview 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 5 Preview

    Step 5 Preview 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

    Step 5 Preview

    Step 5 Preview 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

    Step 5 Preview is not ranked on the public lane for coding, so no winner is named for coding.

    Confidence: limited

  • Agentic work

    Tool use, computer use, and multi-step task completion

    Not enough matched evidence

    Step 5 Preview is not ranked on the public lane for agentic, so no winner is named for agentic.

    Confidence: limited

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.

67.9Kimi K3—Step 5 Preview

Not comparable · BenchAlign

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.

Open the coding leaderboardShortlist models for this job

What is actually comparable

Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.

Shared results
14
Kimi K3 only
30
Step 5 Preview only
8
Like-for-like categories
0 / 8

1 category rests on Estimated evidence or different evidence sets. Those rows remain visible for coverage context but do not name a winner.

Category results, on a stated basis

Each row shows the public-lane category score for both models: the BenchAlign 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

Directional only
Kimi K3
89.4
#1/49
Step 5 Preview
61.2
#28/49
Basis
Provisional lane · 3 vs 2 weighted rows
Reading
Directional only

Agentic

Not comparable
Kimi K3
72.1
Supported · #4/154
Step 5 Preview
Not ranked
Basis
BenchAlign lane · 12 vs 11 public rows
Reading
Not comparable

Coding

Not comparable
Kimi K3
67.9
Supported · #9/156
Step 5 Preview
Not ranked
Basis
BenchAlign lane · 13 vs 6 public rows
Reading
Not comparable

Reasoning

Not comparable
Kimi K3
78.5
#3/19
Step 5 Preview
78.5
Unranked · 2 rankable rows
Basis
Provisional lane · 0 vs 0 weighted rows
Reading
Not comparable

Knowledge

Not comparable
Kimi K3
72.1
Supported · #8/186
Step 5 Preview
Not ranked
Basis
BenchAlign lane · 6 vs 2 public rows
Reading
Not comparable

Multilingual

Not comparable
Kimi K3
Not ranked
Step 5 Preview
Not ranked
Basis
Provisional lane · 0 vs 0 weighted rows
Reading
Not comparable

Instruction following

Not comparable
Kimi K3
Not ranked
Step 5 Preview
Not ranked
Basis
Provisional lane · 0 vs 0 weighted rows
Reading
Not comparable

Math

Not comparable
Kimi K3
Not ranked
Step 5 Preview
Not ranked
Basis
Provisional lane · 0 vs 0 weighted rows
Reading
Not comparable

Ranks count the models scored in each category’s lane, so the agentic, coding, and knowledge denominators (BenchAlign) differ from the provisional-lane categories. Unranked scores sit on the lane’s scale but never name a winner. Like-for-like rows within 0.5 points read as a practical tie.

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.

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
Step 5 Preview
$0.00235
Fits in one request

Step 5 Preview has the lower modeled cost

Costs use the listed standard API rates.

Repository review

50K fresh input + 3K output tokens

Kimi K3
$0.195
Fits in one request
Step 5 Preview
$0.0581
Fits in one request

Step 5 Preview 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 K3
$0.27
Fits in one request
Step 5 Preview
$0.057
Fits in one request

Step 5 Preview has the lower modeled cost

Costs use the listed standard API rates.

Specification differences

Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.

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

Step 5 Preview

$0.05 per 1M cached input tokens

StepFun pricing and rate limits

Reasoning profile

Kimi K3

Reasoning

Step 5 Preview

Reasoning

Weight access

Kimi K3

Pending

Step 5 Preview

Pending

License

Kimi K3

Pending

Step 5 Preview

Pending

Release date

Kimi K3

2026-07-16

Step 5 Preview

2026-09-20

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
At least one model is not scored in the current public ranking lane, so the page does not name an overall quality winner.
Workload cost
Repository review: $0.195 vs $0.0581. Cache-heavy agent loop: $0.27 vs $0.057.
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 evidence52 rows

Agentic

  • Terminal-Bench 2.0

    Kimi K388.3%
    Source
    Step 5 Preview—

    Not directly comparable

  • BrowseComp

    Kimi K391.2%
    Source
    Step 5 Preview88.7%
    Source

    Kimi K3 leads this result

  • DeepSearchQA

    Kimi K395.0%
    Source
    Step 5 Preview—

    Not directly comparable

  • Toolathlon-Verified

    Kimi K373.2%
    Source
    Step 5 Preview74.1%
    Source

    Step 5 Preview leads this result

  • MCP Atlas

    Kimi K384.2%
    Source
    Step 5 Preview85.6%
    Source

    Step 5 Preview leads this result

  • AutomationBench

    Kimi K330.8%
    Source
    Step 5 Preview44.0%
    Source

    Step 5 Preview leads this result

  • JobBench

    Kimi K352.9%
    Source
    Step 5 Preview59.0%
    Source

    Step 5 Preview leads this result

  • APEX-Agents

    Kimi K337.6%
    Source
    Step 5 Preview37.8%
    Source

    Step 5 Preview leads this result

  • SpreadsheetBench 2

    Kimi K334.8%
    Source
    Step 5 Preview—

    Not directly comparable

  • DECK-Bench

    Kimi K373.5%
    Source
    Step 5 Preview—

    Not directly comparable

  • Terminal-Bench 2.1 (Vals)

    Kimi K380.9%
    Source
    Step 5 Preview—

    Not directly comparable

  • ApprenticeBench

    Kimi K318%
    Source
    Step 5 Preview—

    Not directly comparable

  • Terminal-Bench 2.1

    Kimi K3—
    Step 5 Preview85.0%
    Source

    Not directly comparable

  • Terminal-Bench 4.0

    Kimi K3—
    Step 5 Preview33.30%
    Source

    Not directly comparable

  • CyberGym

    Kimi K3—
    Step 5 Preview84.7%
    Source

    Not directly comparable

  • DRACO

    Kimi K3—
    Step 5 Preview83.3%
    Source

    Not directly comparable

  • Agents' Last Exam

    Kimi K3—
    Step 5 Preview29.5%
    Source

    Not directly comparable

Coding

  • DeepSWE

    Kimi K367.5%
    Source
    Step 5 Preview67.7%
    Source

    Step 5 Preview leads this result

  • cursorBench32

    Kimi K360.8%
    Source
    Step 5 Preview—

    Not directly comparable

  • FrontierSWE

    Kimi K381.2%
    Source
    Step 5 Preview—

    Not directly comparable

  • ProgramBench

    Kimi K377.8%
    Source
    Step 5 Preview80.5%
    Source

    Step 5 Preview leads this result

  • Kimi Code Bench v2

    Kimi K372.9%
    Source
    Step 5 Preview—

    Not directly comparable

  • sweMarathon

    Kimi K342%
    Source
    Step 5 Preview72.7%
    Source

    Step 5 Preview leads this result

  • PostTrain Bench

    Kimi K336.6%
    Source
    Step 5 Preview—

    Not directly comparable

  • MLS-Bench Lite

    Kimi K348.3%
    Source
    Step 5 Preview40.5%
    Source

    Kimi K3 leads this result

  • VulcanBench v3

    Kimi K373.7%
    Source
    Step 5 Preview—

    Not directly comparable

  • OpenHarmony Bench

    Kimi K357.3%
    Source
    Step 5 Preview—

    Not directly comparable

  • FrontierSWE v2

    Kimi K325.9%
    Source
    Step 5 Preview—

    Not directly comparable

  • LiveCodeBench (Vals)

    Kimi K387.2%
    Source
    Step 5 Preview—

    Not directly comparable

  • SWE-bench (Vals)

    Kimi K393.4%
    Source
    Step 5 Preview—

    Not directly comparable

  • Terminal-Bench 2.1

    Kimi K3—
    Step 5 Preview85.0%
    Source

    Not directly comparable

  • SciCode

    Kimi K3—
    Step 5 Preview58.9%
    Source

    Not directly comparable

Reasoning

  • CritPt

    Kimi K3—
    Step 5 Preview20.9%
    Source

    Not directly comparable

Multimodal

  • OfficeQA Pro

    Kimi K363.3%
    Source
    Step 5 Preview60.3%
    Source

    Kimi K3 leads this result

  • MMMU-Pro

    Kimi K381.6%
    Source
    Step 5 Preview76%
    Source

    Kimi K3 leads this result

  • MMMU-Pro w/ Python

    Kimi K383.4%
    Source
    Step 5 Preview—

    Not directly comparable

  • CharXiv w/o tools

    Kimi K384.8%
    Source
    Step 5 Preview—

    Not directly comparable

  • CharXiv

    Kimi K391.3%
    Source
    Step 5 Preview—

    Not directly comparable

  • MathVision

    Kimi K394.3%
    Source
    Step 5 Preview—

    Not directly comparable

  • MathVision w/ Python

    Kimi K397.8%
    Source
    Step 5 Preview—

    Not directly comparable

  • BabyVision w/ Python

    Kimi K385.7%
    Source
    Step 5 Preview—

    Not directly comparable

  • ZeroBench

    Kimi K323.0%
    Source
    Step 5 Preview—

    Not directly comparable

  • ZeroBench w/ Python

    Kimi K341.0%
    Source
    Step 5 Preview—

    Not directly comparable

  • WorldVQA ForceAnswer

    Kimi K351.0%
    Source
    Step 5 Preview—

    Not directly comparable

  • OmniDocBench

    Kimi K391.1%
    Source
    Step 5 Preview—

    Not directly comparable

  • PerceptionBench

    Kimi K358.5%
    Source
    Step 5 Preview—

    Not directly comparable

Knowledge

  • GPQA

    Kimi K393.5%
    Source
    Step 5 Preview—

    Not directly comparable

  • GPQA-D

    Kimi K393.5%
    Source
    Step 5 Preview93.5%
    Source

    Tie

  • HLE

    Kimi K356%
    Source
    Step 5 Preview46.5%
    Source

    Kimi K3 leads this result

  • HLE w/o tools

    Kimi K343.5%
    Source
    Step 5 Preview—

    Not directly comparable

  • GPQA Diamond (Vals)

    Kimi K392.9%
    Source
    Step 5 Preview—

    Not directly comparable

  • MMLU-Pro (Vals)

    Kimi K388.0%
    Source
    Step 5 Preview—

    Not directly comparable

Questions

Which is better, Kimi K3 or Step 5 Preview?

At least one model is not scored in the current public ranking lane, so the page does not name an overall quality winner. The page therefore keeps the decision tied to the specific documented workload.

Which is better for coding, Kimi K3 or Step 5 Preview?

Step 5 Preview is not ranked on the public lane for coding, so no winner is named for coding.

Which is better for agentic tasks, Kimi K3 or Step 5 Preview?

Step 5 Preview is not ranked on the public lane for agentic tasks, so no winner is named for agentic tasks.

Which costs less, Kimi K3 or Step 5 Preview?

For the stated presets, chat costs $0.0105 on Kimi K3 and $0.00235 on Step 5 Preview; repository review costs $0.195 and $0.0581; the cache-heavy agent loop costs $0.27 and $0.057. Costs use the listed standard API rates.

Which has the larger context window, Kimi K3 or Step 5 Preview?

Kimi K3 has the larger documented context window: 1.05M, compared with 1M.

Related comparisons

Compare API pricing Read the methodology Open the model selector

Last updated September 21, 2026

Watch Kimi K3 vs Step 5 Preview

One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.

Read a sample issue

Join 2,000+ readers.