Skip to main content
BenchLM
Data

MiniMax M2.5 vs MiniMax M3

Updated November 13, 2026. Rank cannot separate these two. Price, access, and your workload decide. Public scores include evidence status and uncertainty.

Share or export
Share on XLinkedInAPI/MCP

Decision reading

The public evidence has no benchmark result shared by both models, so it does not support a quality verdict. Use the documented cost, context, and runtime rows instead. 0 results are shared. Category rows resting on Estimated evidence or different benchmark sets are marked directional and do not name a winner.

Model A
MiniMax logo
MiniMax M2.5

MiniMax

49.3/100

Supported · Public rank #95

90% interval 39.8–58.8

Model B
MiniMax logo
MiniMax M3

MiniMax

49.41/100

Supported · Public rank #94

90% interval 34.3–64.5

API price per 1M tokens

MiniMax M2.5: $0.30 input / $1.20 output

MiniMax M3: $0.30 input / $1.20 output

Context window

MiniMax M2.5: 128K

MiniMax M3: 1M

Released

MiniMax M2.5: February 12, 2026

MiniMax M3: June 1, 2026

Shared results
0
MiniMax M2.5 only
1
MiniMax M3 only
28
Like-for-like categories
0 / 8
Supported: MiniMax M2.5 and MiniMax M3How the comparison works

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
Show secondary and unsupported calls

  • Coding work

    Code generation, repair, and software-engineering tasks

    Not enough matched evidence

    MiniMax M2.5 is scored on Estimated evidence for coding, so the reading is directional rather than like-for-like.

    Confidence: limited

  • Agentic work

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

    Not enough matched evidence

    MiniMax M2.5 is scored on Estimated evidence for agentic, so the reading is directional rather than like-for-like.

    Confidence: limited

  • Chat turn cost

    1K fresh input + 500 output tokens

    No clear pick

    A complete comparable API-rate estimate is not available for both models.

    Confidence: listed-rates

  • Cache-heavy agent loop cost

    200K cached + 20K fresh input + 10K output tokens

    Not enough matched evidence

    The page does not recommend a cost winner because at least one model cannot fit the stated workload in one request. MiniMax M2.5 does not fit this workload in one request. MiniMax M2.5 has no published cached-input rate, so cached tokens use its listed input rate.

    Confidence: rate-fallback

  • Repository review cost

    50K fresh input + 3K output tokens

    No clear pick

    A complete comparable API-rate estimate is not available for both models.

    Confidence: listed-rates

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.

34.9MiniMax M2.537.9MiniMax M3

Directional only · BenchAlign v5.8

MiniMax M3 has the higher coding point estimate. Conditional score ranges do not establish rank confidence.

Coding scores combine specific tasks and setups. Match the editor, harness, and effort to your workflow.

Open the coding leaderboardShortlist models for this job
Same basis rules as the category table below

Or describe your own job

The LLM Selector reads what you need a model for and returns a shortlist, ranked in the same order the job pages publish.

Describe the task and constraints. This selector recommends models; it cannot carry out the task or access your files and accounts.

We keep what you type here, drafts included, for 180 days, and after that only the job and requirements read from it. Nothing is tied to your IP address or account. Leave out names and other personal details. Privacy

What is actually comparable

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

2 categories rest on Estimated evidence or different evidence sets. Those rows remain visible for coverage context but do not name a winner.

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.

A shared-evidence shape is not available.

BenchLM does not draw a radar or infer missing axes when the matched evidence is too sparse.

Category results, on a stated basis

Each row shows the public-lane category score for both models: the BenchAlign v5.8 lane for agentic, coding, reasoning, multimodal, 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.

Agentic

Directional only
MiniMax M2.5
29.7
Estimated · #107/135
MiniMax M3
38.5
Supported · #75/135
Basis
BenchAlign v5.8 lane · 0 vs 10 public rows
Reading
Directional only

Coding

Directional only
MiniMax M2.5
34.9
Estimated · #93/162
MiniMax M3
37.9
Supported · #74/162
Basis
BenchAlign v5.8 lane · 1 vs 10 public rows
Reading
Directional only

Reasoning

Not comparable
MiniMax M2.5
Not ranked
MiniMax M3
38.7
Estimated · #89/160
Basis
BenchAlign v5.8 lane · 0 vs 0 public rows
Reading
Not comparable

Multimodal

Not comparable
MiniMax M2.5
Not ranked
MiniMax M3
53.7
Estimated · #49/67
Basis
BenchAlign v5.8 lane · 0 vs 5 public rows
Reading
Not comparable

Knowledge

Not comparable
MiniMax M2.5
Not ranked
MiniMax M3
54.9
Supported · #61/180
Basis
BenchAlign v5.8 lane · 0 vs 2 public rows
Reading
Not comparable

Multilingual

Not comparable
MiniMax M2.5
Not ranked
MiniMax M3
Not ranked
Basis
Provisional lane · 0 vs 0 weighted rows
Reading
Not comparable

Instruction following

Not comparable
MiniMax M2.5
Not ranked
MiniMax M3
93.8
#6/157
Basis
Provisional lane · 0 vs 0 weighted rows
Reading
Not comparable

Math

Not comparable
MiniMax M2.5
Not ranked
MiniMax M3
Not ranked
Basis
Provisional lane · 0 vs 1 weighted rows
Reading
Not comparable

Ranks count the models scored in each category’s lane, so the agentic, coding, reasoning, multimodal, and knowledge denominators (BenchAlign v5.8) differ from the provisional-lane categories. Unranked scores sit on the provisional lane’s scale but never name a winner. Like-for-like rows within 0.5 points read as a practical tie.

Supported evidence per lane · bars run 0–100Methodology

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

MiniMax M2.5
$0.0009
Fits in one request
MiniMax M3
$0.0009
Fits in one request

Modeled costs are equal

Costs use the listed standard API rates.

Repository review

50K fresh input + 3K output tokens

MiniMax M2.5
$0.0186
Fits in one request
MiniMax M3
$0.0186
Fits in one request

Modeled costs are equal

Costs use the listed standard API rates.

Cache-heavy agent loop

200K cached + 20K fresh input + 10K output tokens

MiniMax M2.5
$0.078
Does not fit in one request
Cached input priced at the published list-input rate
MiniMax M3
$0.03
Fits in one request

MiniMax M2.5 does not fit this workload in one request. MiniMax M2.5 has no published cached-input rate, so cached tokens use its listed input rate.

Your workload

Enter a month of tokens to price it at both models’ listed API rates.

MiniMax M2.5
$5.40
per month
MiniMax M3
$5.40
per month

Standard listed rates, no batch or volume discounts.

Cached input falls back to the list input rate only where a cached rate is unpublished

Price against score, among all priced models

Overall score against output price per 1M tokens, one model per family. Pricing as of November 12, 2026.

AMiniMax M2.5BMiniMax M3

20406080100$0.10$0.50$1$5$10$50$100Output price, $ per 1M tokens (log)Overall score
  • 7 models score at least as high as MiniMax M2.5 for the same price or less. The nearest is DeepSeek V4.1 Flash 1017 (score 64.5, also $1.20 per 1M output tokens).
  • 6 models score at least as high as MiniMax M3 for the same price or less. The nearest is DeepSeek V4.1 Flash 1017 (score 64.5, also $1.20 per 1M output tokens).

Gray dots are the other priced models. The dark step line is the efficiency frontier: for a model under it, another model costs no more and scores at least as high.

See MiniMax M2.5 on the full price vs performance chart

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.

MiniMax M2.5

128K

MiniMax M3

1M

API model ID

MiniMax M2.5

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.

MiniMax M2.5

Not published

MiniMax M3

$0.06 per 1M cached input tokens

Provider availability

MiniMax M2.5

Not sourced

MiniMax M3

Not sourced

Reasoning profile

MiniMax M2.5

Non-Reasoning

MiniMax M3

Non-Reasoning

Weight access

MiniMax M2.5

Proprietary

MiniMax M3

Open Weight

License

MiniMax M2.5

Proprietary

MiniMax M3

Open Weight

Release date

MiniMax M2.5

2026-02-12

MiniMax M3

2026-06-01

If you already use one of these models

Deployment change
Both entries list MiniMax as the provider. Confirm endpoint, model ID, limits, and feature support before switching.
Quality signal
The public evidence has no benchmark result shared by both models, so it does not support a quality verdict.
Workload cost
Repository review: $0.0186 vs $0.0186. Cache-heavy agent loop: $0.078 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.

Questions

Which is better, MiniMax M2.5 or MiniMax M3?

The public evidence has no benchmark result shared by both models, so it does not support a quality verdict. The page therefore keeps the decision tied to the specific documented workload.

Which is better for coding, MiniMax M2.5 or MiniMax M3?

MiniMax M3 scores higher for coding on the public lane, 37.9 to 34.9. MiniMax M2.5 is scored on Estimated evidence for coding, so the reading is directional rather than like-for-like. BenchLM does not name a winner for coding from a directional reading; read the shared benchmark rows directly and test the models on the same task set.

Which is better for agentic tasks, MiniMax M2.5 or MiniMax M3?

MiniMax M3 scores higher for agentic tasks on the public lane, 38.5 to 29.7. MiniMax M2.5 is scored on Estimated evidence for agentic tasks, so the reading is directional rather than like-for-like. BenchLM does not name a winner for agentic tasks from a directional reading; read the shared benchmark rows directly and test the models on the same task set.

Which costs less, MiniMax M2.5 or MiniMax M3?

For the stated presets, chat costs $0.0009 on MiniMax M2.5 and $0.0009 on MiniMax M3; repository review costs $0.0186 and $0.0186; the cache-heavy agent loop costs $0.078 and $0.03. MiniMax M2.5 does not fit this workload in one request. MiniMax M2.5 has no published cached-input rate, so cached tokens use its listed input rate.

Which has the larger context window, MiniMax M2.5 or MiniMax M3?

MiniMax M3 has the larger documented context window: 1M, compared with 128K.

Self-host vs API cost

Estimates at 50,000 req/day · 1000 tokens/req average.

MiniMax M2.5
API / mo$1,125
Self-host / moNot listed
Break-even—
Proprietary model — self-hosting not applicable.
MiniMax M3
API / mo$1,125
Self-host / mo$13,981
Break-evenNot reached at 70% load
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 evidence29 rows

Agentic

  • Terminal-Bench 2.1

    MiniMax M2.5—
    MiniMax M366.0%
    Source

    Not directly comparable

  • BrowseComp

    MiniMax M2.5—
    MiniMax M383.5%
    Source

    Not directly comparable

  • OSWorld-Verified

    MiniMax M2.5—
    MiniMax M370.1%
    Source

    Not directly comparable

  • MCP Atlas

    MiniMax M2.5—
    MiniMax M374.2%
    Source

    Not directly comparable

  • Claw-Eval

    MiniMax M2.5—
    MiniMax M374.5%
    Source

    Not directly comparable

  • BankerToolBench

    MiniMax M2.5—
    MiniMax M376.1%
    Source

    Not directly comparable

  • ResearchClawBench

    MiniMax M2.5—
    MiniMax M319.8%
    Source

    Not directly comparable

  • OSWorld 2.0

    MiniMax M2.5—
    MiniMax M34.6%
    Source

    Not directly comparable

  • Terminal-Bench 2.1 (Vals)

    MiniMax M2.5—
    MiniMax M353.6%
    Source

    Not directly comparable

  • MazeBench

    MiniMax M2.5—
    MiniMax M30%
    Source

    Not directly comparable

Coding

  • Vibe Code Bench

    MiniMax M2.514.85%
    Source
    MiniMax M3—

    Not directly comparable

  • SWE-bench Verified

    MiniMax M2.5—
    MiniMax M380.5%
    Source

    Not directly comparable

  • SWE-bench Pro

    MiniMax M2.5—
    MiniMax M359%
    Source

    Not directly comparable

  • Terminal-Bench 2.1

    MiniMax M2.5—
    MiniMax M366.0%
    Source

    Not directly comparable

  • NL2Repo

    MiniMax M2.5—
    MiniMax M342.1%
    Source

    Not directly comparable

  • VIBE V2

    MiniMax M2.5—
    MiniMax M350.1%
    Source

    Not directly comparable

  • SVG-Bench

    MiniMax M2.5—
    MiniMax M363.7%
    Source

    Not directly comparable

  • KernelBench Hard

    MiniMax M2.5—
    MiniMax M328.8%
    Source

    Not directly comparable

  • OpenHarmony Bench v1.0

    MiniMax M2.5—
    MiniMax M348.4%
    Source

    Not directly comparable

  • LiveCodeBench (Vals)

    MiniMax M2.5—
    MiniMax M382.2%
    Source

    Not directly comparable

  • SWE-bench (Vals)

    MiniMax M2.5—
    MiniMax M375.0%
    Source

    Not directly comparable

Multimodal

  • OfficeQA Pro

    MiniMax M2.5—
    MiniMax M345.1%
    Source

    Not directly comparable

  • OmniDocBench 1.5

    MiniMax M2.5—
    MiniMax M391.6%
    Source

    Not directly comparable

  • MMMU-Pro

    MiniMax M2.5—
    MiniMax M378.1%
    Source

    Not directly comparable

  • VideoMMMU

    MiniMax M2.5—
    MiniMax M384.6%
    Source

    Not directly comparable

  • Video-MME (with subtitle)

    MiniMax M2.5—
    MiniMax M385.4%
    Source

    Not directly comparable

Knowledge

  • GPQA Diamond (Vals)

    MiniMax M2.5—
    MiniMax M392.7%
    Source

    Not directly comparable

  • MMLU-Pro (Vals)

    MiniMax M2.5—
    MiniMax M384.2%
    Source

    Not directly comparable

Math

  • USAMO 2026

    MiniMax M2.5—
    MiniMax M385.7%
    Source

    Not directly comparable

29 public results · 0 shared

Related comparisons

Compare API pricingRead the methodologyOpen the model selector

Watch MiniMax M2.5 vs MiniMax M3

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

Read a sample issue

Join 7,000+ subscribers.

Last updated November 13, 2026