Skip to main content
Radar

Five or fewer confirmed AI changes, with original sources, on mornings when something changed.A free source-linked morning brief.

Start the free Radar Brief
BenchLM
Moonshot AI logo
Model A
Kimi K2.5

Moonshot AI

59.14/100

Supported · Public rank #80

90% interval 50.9–67.4

Kimi K2.5 vs MiniMax M3

Updated August 30, 2026. Public scores include evidence status and uncertainty. They are not guarantees for a specific workload.

MiniMax logo
Model B
MiniMax M3

MiniMax

68.73/100

Supported · Public rank #21

90% interval 63.5–74.0

Decision reading

MiniMax M3 has the higher public score estimate, 68.73 versus 59.14, but the 90% score intervals overlap. Treat that as a lead, not a settled winner.

9 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.

  • Long documents

    Prompts that approach the documented context limit

    MiniMax M3

    MiniMax M3 has the larger documented context window.

    Confidence: documented

  • Chat turn cost

    1K fresh input + 500 output tokens

    MiniMax M3

    MiniMax M3 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

    MiniMax M3

    MiniMax M3 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.

    Confidence: rate-fallback

Show secondary and unsupported calls

  • Repository review cost

    50K fresh input + 3K output tokens

    MiniMax M3

    MiniMax M3 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

  • Agentic work

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

    Not enough matched evidence

    The category averages use different weighted benchmark sets, so they are directional rather than like-for-like.

    Confidence: limited

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
9
Kimi K2.5 only
36
MiniMax M3 only
13
Like-for-like categories
0 / 8

3 categories use 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.

Agentic

Directional only
Kimi K2.5
55.0
MiniMax M3
72.3
Weighted basis
2 vs 3 rows
Reading
Directional only

Coding

Directional only
Kimi K2.5
59.4
MiniMax M3
72.2
Weighted basis
4 vs 2 rows
Reading
Directional only

Multimodal

Directional only
Kimi K2.5
78.5
MiniMax M3
64.9
Weighted basis
1 vs 2 rows
Reading
Directional only

Reasoning

Not comparable
Kimi K2.5
61.0
MiniMax M3
Not measured
Weighted basis
1 vs 0 rows
Reading
Not comparable

Knowledge

Not comparable
Kimi K2.5
56.9
MiniMax M3
Not measured
Weighted basis
4 vs 0 rows
Reading
Not comparable

Math

Not comparable
Kimi K2.5
60.6
MiniMax M3
85.7
Weighted basis
4 vs 1 rows
Reading
Not comparable

Multilingual

Not comparable
Kimi K2.5
82.3
MiniMax M3
Not measured
Weighted basis
1 vs 0 rows
Reading
Not comparable

Instruction following

Not comparable
Kimi K2.5
93.9
MiniMax M3
Not measured
Weighted basis
1 vs 0 rows
Reading
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.

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
MiniMax M3
$0.0009
Fits in one request

MiniMax M3 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
MiniMax M3
$0.0186
Fits in one request

MiniMax M3 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
MiniMax M3
$0.03
Fits in one request

MiniMax M3 has the lower modeled cost

Kimi K2.5 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.

Context window

Maximum documented context; output-token limits may be lower.

Kimi K2.5

256K

MiniMax M3

1M

API model ID

Kimi K2.5

Not sourced

MiniMax M3

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

MiniMax M3

$0.06 per 1M cached input tokens

Documented inputs

Kimi K2.5

Not sourced

MiniMax M3

Not sourced

Documented outputs

Kimi K2.5

Not sourced

MiniMax M3

Not sourced

Provider availability

Kimi K2.5

Not sourced

MiniMax M3

Not sourced

Reasoning profile

Kimi K2.5

Non-Reasoning

MiniMax M3

Non-Reasoning

Weight access

Kimi K2.5

Open Weight

MiniMax M3

Open Weight

License

Kimi K2.5

Open Weight

MiniMax M3

Open Weight

Release date

Kimi K2.5

2026-02-01

MiniMax M3

2026-06-01

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
MiniMax M3 has the higher public score estimate, 68.73 versus 59.14, but the 90% score intervals overlap.
Workload cost
Repository review: $0.039 vs $0.0186. Cache-heavy agent loop: $0.162 vs $0.03.
Context tradeoff
MiniMax M3 has the larger documented window (1M).

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
MiniMax M3
API / mo$1,125
Self-host / moNot listed
Break-even—
Proprietary model — self-hosting not applicable.
Model the full break-even

Benchmark evidence

The full public result ledger is available for audit without forcing a wide desktop table onto a phone.

Browse raw public benchmark evidence58 rows

Agentic

  • Terminal-Bench 2.0

    Kimi K2.550.8%
    Source
    MiniMax M366%
    Source

    MiniMax M3 leads this result

  • BrowseComp

    Kimi K2.560.6%
    Source
    MiniMax M383.5%
    Source

    MiniMax M3 leads this result

  • Claw-Eval

    Kimi K2.552.3%
    Source
    MiniMax M374.5%
    Source

    MiniMax M3 leads this result

  • QwenClawBench

    Kimi K2.554.3%
    Source
    MiniMax M3—

    Not directly comparable

  • τ³-bench results

    Kimi K2.565.7%
    Source
    MiniMax M3—

    Not directly comparable

  • DeepSearchQA

    Kimi K2.577.1%
    Source
    MiniMax M3—

    Not directly comparable

  • DeepPlanning

    Kimi K2.514.4%
    Source
    MiniMax M3—

    Not directly comparable

  • Toolathlon

    Kimi K2.527.8%
    Source
    MiniMax M3—

    Not directly comparable

  • MCP Atlas

    Kimi K2.529.5%
    Source
    MiniMax M374.2%
    Source

    MiniMax M3 leads this result

  • MCP-Tasks

    Kimi K2.559.1%
    Source
    MiniMax M3—

    Not directly comparable

  • WideResearch

    Kimi K2.572.7%
    Source
    MiniMax M3—

    Not directly comparable

  • Gert Labs

    Kimi K2.545.88%
    Source
    MiniMax M3—

    Not directly comparable

  • ResearchClawBench

    Shared source
    Kimi K2.514.0%
    MiniMax M319.8%

    MiniMax M3 leads this result

  • JobBench

    Kimi K2.58.7%
    Source
    MiniMax M3—

    Not directly comparable

  • OSWorld-Verified

    Kimi K2.5—
    MiniMax M370.1%
    Source

    Not directly comparable

  • BankerToolBench

    Kimi K2.5—
    MiniMax M376.1%
    Source

    Not directly comparable

  • OSWorld 2.0

    Kimi K2.5—
    MiniMax M34.6%
    Source

    Not directly comparable

Coding

  • SWE-bench Verified

    Kimi K2.576.8%
    Source
    MiniMax M380.5%
    Source

    MiniMax M3 leads this result

  • SWE-bench Verified*

    Kimi K2.570.8%
    Source
    MiniMax M3—

    Not directly comparable

  • LiveCodeBench v6

    Kimi K2.585.0%
    Source
    MiniMax M3—

    Not directly comparable

  • SWE-bench Pro

    Kimi K2.550.7%
    Source
    MiniMax M359%
    Source

    MiniMax M3 leads this result

  • SWE Multilingual

    Kimi K2.573%
    Source
    MiniMax M3—

    Not directly comparable

  • SWE-Rebench

    Kimi K2.558.5%
    Source
    MiniMax M3—

    Not directly comparable

  • React Native Evals

    Kimi K2.577.2%
    Source
    MiniMax M3—

    Not directly comparable

  • SciCode

    Kimi K2.548.7%
    Source
    MiniMax M3—

    Not directly comparable

  • Terminal-Bench 2.0

    Kimi K2.5—
    MiniMax M366.0%
    Source

    Not directly comparable

  • NL2Repo

    Kimi K2.5—
    MiniMax M342.1%
    Source

    Not directly comparable

  • VIBE V2

    Kimi K2.5—
    MiniMax M350.1%
    Source

    Not directly comparable

  • SVG-Bench

    Kimi K2.5—
    MiniMax M363.7%
    Source

    Not directly comparable

  • KernelBench Hard

    Kimi K2.5—
    MiniMax M328.8%
    Source

    Not directly comparable

  • OpenHarmony Bench

    Kimi K2.5—
    MiniMax M348.4%
    Source

    Not directly comparable

Reasoning

  • LongBench v2

    Kimi K2.561%
    Source
    MiniMax M3—

    Not directly comparable

Knowledge

  • GPQA

    Kimi K2.587.6%
    Source
    MiniMax M3—

    Not directly comparable

  • GPQA-D

    Kimi K2.587.6%
    Source
    MiniMax M3—

    Not directly comparable

  • SuperGPQA

    Kimi K2.569.2%
    Source
    MiniMax M3—

    Not directly comparable

  • MMLU-Pro

    Kimi K2.587.1%
    Source
    MiniMax M3—

    Not directly comparable

  • MMLU-Pro (Arcee)

    Kimi K2.587.1%
    Source
    MiniMax M3—

    Not directly comparable

  • HLE

    Kimi K2.530.1%
    Source
    MiniMax M3—

    Not directly comparable

Math

  • AIME 2025

    Kimi K2.596.1%
    Source
    MiniMax M3—

    Not directly comparable

  • AIME26

    Kimi K2.595.8%
    Source
    MiniMax M3—

    Not directly comparable

  • AIME25 (Arcee)

    Kimi K2.596.3%
    Source
    MiniMax M3—

    Not directly comparable

  • HMMT Feb 2025

    Kimi K2.595.4%
    Source
    MiniMax M3—

    Not directly comparable

  • HMMT Nov 2025

    Kimi K2.591.1%
    Source
    MiniMax M3—

    Not directly comparable

  • HMMT Feb 2026

    Kimi K2.587.1%
    Source
    MiniMax M3—

    Not directly comparable

  • MMAnswerBench

    Kimi K2.581.8%
    Source
    MiniMax M3—

    Not directly comparable

  • FrontierMath v2 (Tiers 1-3)

    Kimi K2.527.900%
    Source
    MiniMax M3—

    Not directly comparable

  • FrontierMath v2 (Tier 4)

    Kimi K2.54.200%
    Source
    MiniMax M3—

    Not directly comparable

  • USAMO 2026

    Kimi K2.5—
    MiniMax M385.7%
    Source

    Not directly comparable

Multilingual

  • MMLU-ProX

    Kimi K2.582.3%
    Source
    MiniMax M3—

    Not directly comparable

  • NOVA-63

    Kimi K2.556.0%
    Source
    MiniMax M3—

    Not directly comparable

Multimodal

  • MMMU-Pro

    Kimi K2.578.5%
    Source
    MiniMax M378.1%
    Source

    Kimi K2.5 leads this result

  • Video-MME

    Kimi K2.587.4%
    Source
    MiniMax M3—

    Not directly comparable

  • MMVU

    Kimi K2.580.4%
    Source
    MiniMax M3—

    Not directly comparable

  • VideoMMMU

    Kimi K2.586.6%
    Source
    MiniMax M384.6%
    Source

    Kimi K2.5 leads this result

  • OfficeQA Pro

    Kimi K2.5—
    MiniMax M345.1%
    Source

    Not directly comparable

  • OmniDocBench 1.5

    Kimi K2.5—
    MiniMax M391.6%
    Source

    Not directly comparable

  • Video-MME (with subtitle)

    Kimi K2.5—
    MiniMax M385.4%
    Source

    Not directly comparable

Instruction following

  • IFEval

    Kimi K2.593.9%
    Source
    MiniMax M3—

    Not directly comparable

Frequently asked questions

Which is better, Kimi K2.5 or MiniMax M3?

MiniMax M3 has the higher public score estimate, 68.73 versus 59.14, 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 MiniMax M3?

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 MiniMax M3?

The current agentic tasks 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 costs less, Kimi K2.5 or MiniMax M3?

For the stated presets, chat costs $0.0021 on Kimi K2.5 and $0.0009 on MiniMax M3; repository review costs $0.039 and $0.0186; the cache-heavy agent loop costs $0.162 and $0.03. Kimi K2.5 has no published cached-input rate, so cached tokens use its listed input rate.

Which has the larger context window, Kimi K2.5 or MiniMax M3?

MiniMax M3 has the larger documented context window: 1M, compared with 256K.

Related comparisons

Compare API pricing Read the methodology Open the model selector

Last updated August 30, 2026

Watch Kimi K2.5 vs MiniMax M3

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

Read a sample issue

Join 2,000+ readers.