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.

See the free Radar Brief
BenchLM
Microsoft logo
Model A
MAI-Thinking-1

Microsoft

52.27/100

Estimated · Public rank #123

90% interval 42.4–62.1

MAI-Thinking-1 vs MiniMax M3

Updated September 4, 2026. Public scores include evidence status and uncertainty. They are not guarantees for a specific workload.

MiniMax logo
Model B
MiniMax M3

MiniMax

63.91/100

Supported · Public rank #51

90% interval 56.5–71.3

Decision reading

MiniMax M3 has the higher public score estimate, 63.91 versus 52.27, but the 90% score intervals overlap. Treat that as a lead, not a settled winner.

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

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

    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

    MAI-Thinking-1 and MiniMax M3 are 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

    MAI-Thinking-1 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

    Not enough matched evidence

    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

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

    Confidence: listed-rates

  • Repository review cost

    50K fresh input + 3K output tokens

    Not enough matched evidence

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

    Confidence: listed-rates

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
4
MAI-Thinking-1 only
10
MiniMax M3 only
23
Like-for-like categories
0 / 8

4 categories rest 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.

Agentic

Directional only
MAI-Thinking-1
51.7
Estimated · #53/151
MiniMax M3
43.2
Supported · #106/151
Basis
BenchAlign lane · 1 vs 9 public rows
Reading
Directional only

Coding

Directional only
MAI-Thinking-1
51.8
Estimated · #60/183
MiniMax M3
50.6
Estimated · #68/183
Basis
BenchAlign lane · 4 vs 10 public rows
Reading
Directional only

Knowledge

Directional only
MAI-Thinking-1
53.3
Estimated · #68/181
MiniMax M3
53.9
Supported · #65/181
Basis
BenchAlign lane · 4 vs 2 public rows
Reading
Directional only

Instruction following

Directional only
MAI-Thinking-1
94.7
#1/120
MiniMax M3
93.5
#4/120
Basis
Provisional lane · 1 vs 0 weighted rows
Reading
Directional only

Reasoning

Not comparable
MAI-Thinking-1
Not ranked
MiniMax M3
78.5
#5/22
Basis
Provisional lane · 0 vs 0 weighted rows
Reading
Not comparable

Math

Not comparable
MAI-Thinking-1
73.4
Unranked · 3 rankable rows
MiniMax M3
Not ranked
Basis
Provisional lane · 2 vs 1 weighted rows
Reading
Not comparable

Multilingual

Not comparable
MAI-Thinking-1
Not ranked
MiniMax M3
Not ranked
Basis
Provisional lane · 0 vs 0 weighted rows
Reading
Not comparable

Multimodal

Not comparable
MAI-Thinking-1
Not ranked
MiniMax M3
52.0
#34/48
Basis
Provisional lane · 0 vs 2 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

MAI-Thinking-1
API rate not published
Fits in one request
MiniMax M3
$0.0009
Fits in one request

MAI-Thinking-1 has no comparable published API token rate.

Repository review

50K fresh input + 3K output tokens

MAI-Thinking-1
API rate not published
Fits in one request
MiniMax M3
$0.0186
Fits in one request

MAI-Thinking-1 has no comparable published API token rate.

Cache-heavy agent loop

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

MAI-Thinking-1
API rate not published
Fits in one request
Cached-input rate unavailable
MiniMax M3
$0.03
Fits in one request

MAI-Thinking-1 has no comparable published API token 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.

MAI-Thinking-1

256K

MiniMax M3

1M

API model ID

MAI-Thinking-1

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.

MAI-Thinking-1

No comparable hosted API rate

MiniMax M3

$0.06 per 1M cached input tokens

Documented inputs

MAI-Thinking-1

Not sourced

MiniMax M3

Not sourced

Documented outputs

MAI-Thinking-1

Not sourced

MiniMax M3

Not sourced

Provider availability

MAI-Thinking-1

Not sourced

MiniMax M3

Not sourced

Reasoning profile

MAI-Thinking-1

Reasoning

MiniMax M3

Non-Reasoning

Weight access

MAI-Thinking-1

Proprietary

MiniMax M3

Open Weight

License

MAI-Thinking-1

Proprietary

MiniMax M3

Open Weight

Release date

MAI-Thinking-1

2026-06-02

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, 63.91 versus 52.27, but the 90% score intervals overlap.
Workload cost
A complete comparable API-rate estimate is not available for both models.
Context tradeoff
MiniMax M3 has the larger documented window (1M).

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 evidence37 rows

Agentic

  • Terminal-Bench 2.0

    MAI-Thinking-146%
    Source
    MiniMax M366%
    Source

    MiniMax M3 leads this result

  • BrowseComp

    MAI-Thinking-1—
    MiniMax M383.5%
    Source

    Not directly comparable

  • OSWorld-Verified

    MAI-Thinking-1—
    MiniMax M370.1%
    Source

    Not directly comparable

  • MCP Atlas

    MAI-Thinking-1—
    MiniMax M374.2%
    Source

    Not directly comparable

  • Claw-Eval

    MAI-Thinking-1—
    MiniMax M374.5%
    Source

    Not directly comparable

  • BankerToolBench

    MAI-Thinking-1—
    MiniMax M376.1%
    Source

    Not directly comparable

  • ResearchClawBench

    MAI-Thinking-1—
    MiniMax M319.8%
    Source

    Not directly comparable

  • OSWorld 2.0

    MAI-Thinking-1—
    MiniMax M34.6%
    Source

    Not directly comparable

  • Terminal-Bench 2.1 (Vals)

    MAI-Thinking-1—
    MiniMax M353.6%
    Source

    Not directly comparable

Coding

  • LiveCodeBench v6

    MAI-Thinking-187.7%
    Source
    MiniMax M3—

    Not directly comparable

  • SWE-bench Verified

    MAI-Thinking-173.5%
    Source
    MiniMax M380.5%
    Source

    MiniMax M3 leads this result

  • SWE-bench Pro

    MAI-Thinking-152.8%
    Source
    MiniMax M359%
    Source

    MiniMax M3 leads this result

  • Terminal-Bench 2.0

    MAI-Thinking-146.0%
    Source
    MiniMax M366.0%
    Source

    MiniMax M3 leads this result

  • NL2Repo

    MAI-Thinking-1—
    MiniMax M342.1%
    Source

    Not directly comparable

  • VIBE V2

    MAI-Thinking-1—
    MiniMax M350.1%
    Source

    Not directly comparable

  • SVG-Bench

    MAI-Thinking-1—
    MiniMax M363.7%
    Source

    Not directly comparable

  • KernelBench Hard

    MAI-Thinking-1—
    MiniMax M328.8%
    Source

    Not directly comparable

  • OpenHarmony Bench

    MAI-Thinking-1—
    MiniMax M348.4%
    Source

    Not directly comparable

  • LiveCodeBench (Vals)

    MAI-Thinking-1—
    MiniMax M382.2%
    Source

    Not directly comparable

  • SWE-bench (Vals)

    MAI-Thinking-1—
    MiniMax M375.0%
    Source

    Not directly comparable

Reasoning

  • Graphwalks BFS 128K

    MAI-Thinking-190%
    Source
    MiniMax M3—

    Not directly comparable

Knowledge

  • GPQA

    MAI-Thinking-184.2%
    Source
    MiniMax M3—

    Not directly comparable

  • GPQA-D

    MAI-Thinking-184.2%
    Source
    MiniMax M3—

    Not directly comparable

  • MMLU-Pro

    MAI-Thinking-185%
    Source
    MiniMax M3—

    Not directly comparable

  • SimpleQA

    MAI-Thinking-131%
    Source
    MiniMax M3—

    Not directly comparable

  • GPQA Diamond (Vals)

    MAI-Thinking-1—
    MiniMax M392.7%
    Source

    Not directly comparable

  • MMLU-Pro (Vals)

    MAI-Thinking-1—
    MiniMax M384.2%
    Source

    Not directly comparable

Math

  • AIME 2025

    MAI-Thinking-197%
    Source
    MiniMax M3—

    Not directly comparable

  • AIME26

    MAI-Thinking-194.5%
    Source
    MiniMax M3—

    Not directly comparable

  • HMMT Feb 2026

    MAI-Thinking-184.9%
    Source
    MiniMax M3—

    Not directly comparable

  • USAMO 2026

    MAI-Thinking-1—
    MiniMax M385.7%
    Source

    Not directly comparable

Multimodal

  • OfficeQA Pro

    MAI-Thinking-1—
    MiniMax M345.1%
    Source

    Not directly comparable

  • OmniDocBench 1.5

    MAI-Thinking-1—
    MiniMax M391.6%
    Source

    Not directly comparable

  • MMMU-Pro

    MAI-Thinking-1—
    MiniMax M378.1%
    Source

    Not directly comparable

  • VideoMMMU

    MAI-Thinking-1—
    MiniMax M384.6%
    Source

    Not directly comparable

  • Video-MME (with subtitle)

    MAI-Thinking-1—
    MiniMax M385.4%
    Source

    Not directly comparable

Instruction following

  • IFBench

    MAI-Thinking-185%
    Source
    MiniMax M3—

    Not directly comparable

Frequently asked questions

Which is better, MAI-Thinking-1 or MiniMax M3?

MiniMax M3 has the higher public score estimate, 63.91 versus 52.27, but the 90% score intervals overlap. The higher estimate is not a decisive winner because the uncertainty ranges overlap.

Which is better for coding, MAI-Thinking-1 or MiniMax M3?

MAI-Thinking-1 scores higher for coding on the public lane, 51.8 to 50.6. MAI-Thinking-1 and MiniMax M3 are 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, MAI-Thinking-1 or MiniMax M3?

MAI-Thinking-1 scores higher for agentic tasks on the public lane, 51.7 to 43.2. MAI-Thinking-1 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, MAI-Thinking-1 or MiniMax M3?

Both models do not have comparable published API token rates, so this page does not name a universal price winner.

Which has the larger context window, MAI-Thinking-1 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 September 4, 2026

Watch MAI-Thinking-1 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.