Updated Jun 2, 2026

Share

Loading...

Compare MiniMax M3

MiniMax M3 vs MiniMax M2.752 shared prompts · Older

MiniMax M3 vs Ox Alpha53 shared prompts · Cheaper

MiniMax M3 vs Gemini 3.8 Flash53 shared prompts · Premium

MiniMax M3 vs Muse Spark 1.353 shared prompts · Premium

MiniMax M3 vs Dots3-Note Preview52 shared prompts · Cheaper

MiniMax M3 vs Muse Spark 1.3 Contributor53 shared prompts · Cheaper

MiniMax M3 vs Gemini 3.7 Flash53 shared prompts

MiniMax M3 vs Grok 4.653 shared prompts · Premium

Alternatives to MiniMax M3

MiniMax M3's competitors have been quietly putting in work.

GPT-6 Astra Proopenai

Muse Spark 1.3 Contributormeta

Gemini 3.8 Flashgoogle

Claude Fable 5.1anthropic

Mercury 2.5 Previewinception

Granite 4.2 8Bibm-granite

Hy4 Previewtencent

Updated Jun 2, 2026

Share

MiniMax M3

Name: MiniMax M3
Brand: minimax
Price: 0.30 USD
Availability: OnlineOnly

MiniMax:

M1

M2.5

M2.7

M3

MiniMax M3 takes text, image and video, returns text, over a 1M token context, for long-horizon agent work, coding and tool use. MiniMax Sparse Attention swaps full attention for KV-block selection, cutting per-token compute at long context to roughly a twentieth of the last generation.

Website OpenRouter HuggingFace Docs API GitHub Paper Blog

Provider: MiniMax
Released: 2026-06-01
Size: LARGE
Free API (OpenRouter): No
Per 1M tokens: $0.30 in · $1.2 out
Rival’s tab: $0.82; 57/57 receipts survived

Free API availability

No free endpoint for MiniMax M3 was found in Rival’s OpenRouter listings. Other providers or chat apps may have separate free offers.

Provider data: 8 Sep 2026OpenRouter usage limits

View API options

Benchmarks

SWE-Bench Pro

59.0%

source ↗

Terminal-Bench 2.1

66.0%

source ↗

SWE-fficiency

34.8%

source ↗

KernelBench Hard

28.8%

source ↗

MCP Atlas

74.2%

source ↗

Get API accessProvider and language code samples

12 hostsInOutContextUptime

CoreWeavefp4$0.23 in$0.96 out262k100% up GMI Cloudfp8$0.24 in$0.96 out1M99.3% up DeepInfrafp8$0.28 in$1.10 out524k99.4% up MiniMaxfp8$0.30 in$1.20 out524k99.6% up Novitafp8$0.30 in$1.20 out1M99.4% up Parasailfp8$0.30 in$1.20 out1M99.3% up

6 more hosts

StreamLakefp8$0.30 in$1.20 out1M99.1% up Together$0.30 in$1.20 out524k99.8% up Venicefp8$0.30 in$1.20 out524k100% up SambaNova$0.60 in$2.40 out1M99.6% up ModelRunfp4$0.75 in$3.00 out1M99.7% up AtlasCloudfp8degraded$0.30 in$1.20 out524k98.6% up

Per million tokens. Prices and uptime via OpenRouter, checked 8 Sep 2026.

Provider

fromimport openai  OpenAI

client = OpenAI(
"https://openrouter.ai/api/v1"    base_url=,
"$OPENROUTER_API_KEY"    api_key=,
)

response = client.chat.completions.create(
"minimax/minimax-m3"    model=,
"role""user""content""Hello!"    messages=[{: , : }],
)
print(response.choices[0].message.content)

Set OPENROUTER_API_KEY with your OpenRouter API key from openrouter.ai/keys.

Also on NVIDIA NIM

Deep analysisPsychometrics, taste index, writing DNA

SubjectiveBench

Taste Index

Across 57 scored outputs

How this is measured →

390.39xThe default

0100headroom →

Taste is judged on an uncapped scale, originality first. The space past 100 is craft today's models rarely reach.

Craft50

Originality33

Plays it safe

share of outputs that are the default answer

33%

Writing DNA

Stylometric Fingerprint

Based on 27 text responses

Tick = global average

Vocabulary Diversity55%

Unique words vs. total words. Higher = richer vocabulary.

Sentence Length23.7 words

Average words per sentence.

Hedging0.52

"Might", "perhaps", "arguably" per 100 words.

Bold Formatting5.2

**Bold** markers per 1,000 characters.

List Usage3.0

Bullet and numbered list items per 1,000 characters.

Section Structure1.21

Markdown headings per 1,000 characters.

Emoji Usage1.15

Emoji per 1,000 characters.

Transitions0.02

"However", "moreover", "furthermore" per 100 words.

Opening Habits

Starts with heading (81%)

Consistency

67%

Across 27 responses

Favorites

Movie

The Shawshank Redemption

19948.7

Album

Thriller

Michael Jackson

Book

Мастер и Маргарита

Михаил Афанасьевич Булгаков

City

Kyoto

Japan

Game

The Legend of Zelda: Ocarina of Time

Adventure, Action4.4

Model Responses

52 outputs · $0.82 tracked across 52 receipts

Keep exploring

COMPARE

MiniMax M3 vs Ox Alpha

Real outputs compared side by side

RANKINGS

Best AI for Complex Reasoning

Sally's brothers, GPT-3 training FLOPs, a document editor syncing on client c...

Loading...

Compare MiniMax M3

MiniMax M3 vs MiniMax M2.752 shared prompts · Older

MiniMax M3 vs Ox Alpha53 shared prompts · Cheaper

MiniMax M3 vs Gemini 3.8 Flash53 shared prompts · Premium

MiniMax M3 vs Muse Spark 1.353 shared prompts · Premium

MiniMax M3 vs Dots3-Note Preview52 shared prompts · Cheaper

MiniMax M3 vs Muse Spark 1.3 Contributor53 shared prompts · Cheaper

MiniMax M3 vs Gemini 3.7 Flash53 shared prompts

MiniMax M3 vs Grok 4.653 shared prompts · Premium

Alternatives to MiniMax M3

MiniMax M3's competitors have been quietly putting in work.

GPT-6 Astra Proopenai

Muse Spark 1.3 Contributormeta

Gemini 3.8 Flashgoogle

Claude Fable 5.1anthropic

Mercury 2.5 Previewinception

Granite 4.2 8Bibm-granite

Hy4 Previewtencent

MiniMax M3

MiniMax:

M1

M2.5

M2.7

M3

MiniMax M3 takes text, image and video, returns text, over a 1M token context, for long-horizon agent work, coding and tool use. MiniMax Sparse Attention swaps full attention for KV-block selection, cutting per-token compute at long context to roughly a twentieth of the last generation.

Website OpenRouter HuggingFace Docs API GitHub Paper Blog

Provider: MiniMax
Released: 2026-06-01
Size: LARGE
Free API (OpenRouter): No
Per 1M tokens: $0.30 in · $1.2 out
Rival’s tab: $0.82; 57/57 receipts survived

Free API availability

No free endpoint for MiniMax M3 was found in Rival’s OpenRouter listings. Other providers or chat apps may have separate free offers.

Provider data: 8 Sep 2026OpenRouter usage limits

View API options

Benchmarks

SWE-Bench Pro

59.0%

source ↗

Terminal-Bench 2.1

66.0%

source ↗

SWE-fficiency

34.8%

source ↗

KernelBench Hard

28.8%

source ↗

MCP Atlas

74.2%

source ↗

Get API accessProvider and language code samples

12 hostsInOutContextUptime

CoreWeavefp4$0.23 in$0.96 out262k100% up GMI Cloudfp8$0.24 in$0.96 out1M99.3% up DeepInfrafp8$0.28 in$1.10 out524k99.4% up MiniMaxfp8$0.30 in$1.20 out524k99.6% up Novitafp8$0.30 in$1.20 out1M99.4% up Parasailfp8$0.30 in$1.20 out1M99.3% up

6 more hosts

StreamLakefp8$0.30 in$1.20 out1M99.1% up Together$0.30 in$1.20 out524k99.8% up Venicefp8$0.30 in$1.20 out524k100% up SambaNova$0.60 in$2.40 out1M99.6% up ModelRunfp4$0.75 in$3.00 out1M99.7% up AtlasCloudfp8degraded$0.30 in$1.20 out524k98.6% up

Per million tokens. Prices and uptime via OpenRouter, checked 8 Sep 2026.

Provider

fromimport openai  OpenAI

client = OpenAI(
"https://openrouter.ai/api/v1"    base_url=,
"$OPENROUTER_API_KEY"    api_key=,
)

response = client.chat.completions.create(
"minimax/minimax-m3"    model=,
"role""user""content""Hello!"    messages=[{: , : }],
)
print(response.choices[0].message.content)

Set OPENROUTER_API_KEY with your OpenRouter API key from openrouter.ai/keys.

Also on NVIDIA NIM

Deep analysisPsychometrics, taste index, writing DNA

SubjectiveBench

Taste Index

Across 57 scored outputs

How this is measured →

390.39xThe default

0100headroom →

Taste is judged on an uncapped scale, originality first. The space past 100 is craft today's models rarely reach.

Craft50

Originality33

Plays it safe

share of outputs that are the default answer

33%

Writing DNA

Stylometric Fingerprint

Based on 27 text responses

Tick = global average

Vocabulary Diversity55%

Unique words vs. total words. Higher = richer vocabulary.

Sentence Length23.7 words

Average words per sentence.

Hedging0.52

"Might", "perhaps", "arguably" per 100 words.

Bold Formatting5.2

**Bold** markers per 1,000 characters.

List Usage3.0

Bullet and numbered list items per 1,000 characters.

Section Structure1.21

Markdown headings per 1,000 characters.

Emoji Usage1.15

Emoji per 1,000 characters.

Transitions0.02

"However", "moreover", "furthermore" per 100 words.

Opening Habits

Starts with heading (81%)

Consistency

67%

Across 27 responses

Favorites

Movie

The Shawshank Redemption

19948.7

Album

Thriller

Michael Jackson

Book

Мастер и Маргарита

Михаил Афанасьевич Булгаков

City

Kyoto

Japan

Game

The Legend of Zelda: Ocarina of Time

Adventure, Action4.4

Model Responses

52 outputs · $0.82 tracked across 52 receipts

Keep exploring

COMPARE

MiniMax M3 vs Ox Alpha

Real outputs compared side by side

RANKINGS

Best AI for Complex Reasoning

Sally's brothers, GPT-3 training FLOPs, a document editor syncing on client c...