MiniMax · Open weights
MiniMax M3 Thinking
MiniMax M3 Thinking is the adaptive-thinking version of MiniMax's open-weights frontier model for coding, agent workflows, tool use, long-context tasks, and native multimodal understanding. MiniMax reports 59.0% on SWE-Bench Pro and 66.0% on Terminal Bench 2.1, with Sparse Attention designed to scale context to 1M.
- Reasoning
- Tools
- Vision
- Structured output
- Input
- Text, Images
- Output
- Text
- Discovery tier
- Open weights
Pricing
USD per 1M tokens- Input
- $0.23
- Output
- $0.96
- Cache read
- $0.05
Reference prices as published by the model provider. API example
Specifications
- Context window
- 512K
- Max output
- 80K
- Average output (7 days)
- 1.9K tokens
- Longer than 90% of 465 models
Side by side
MiniMax M3 Thinking compared
Specifications, prices, speed and benchmark scores next to each other. Benchmarks below show every compared model too.
| MiniMax M3 ThinkingMiniMaxThis model | Claude Sonnet 4.6 ThinkingAnthropicRemove | |
|---|---|---|
| Provider | MiniMax | Anthropic |
| Discovery tier | Open weights | Frontier |
| Added | Jun 1, 2026 | Feb 17, 2026 |
| Limits | ||
| Context window | 512K | 1MBest |
| Max output | 80K | 128KBest |
| Average output | 1.9K tokens | 598 tokens |
| Pricing, USD per 1M tokens | ||
| Input | $0.23Best | $3.00 |
| Output | $0.96Best | $15.00 |
| Cache read | $0.05Best | $0.30 |
| Performance | ||
| Throughput | 218.4 tok/sBest | 174.4 tok/s |
| Time to first token | 7.2 sBest | 15.2 s |
| Request success (24h) | 99.9% | 100.0%Best |
| Benchmark indexes | ||
| Intelligence Index | 29.2 | 30.1Best |
| Coding Index | 58.6 | 63.0Best |
| Capabilities | ||
| Reasoning | Yes | Yes |
| Tools | Yes | Yes |
| Vision | Yes | Yes |
| Structured output | Yes | Yes |
| PDF input | No | Yes |
| Reasoning effort | — | low, medium, high, xhigh, max |
Public evaluations
Benchmarks
Independent evaluation results, with how each score ranks among the models in the Noumen catalog.
- MiniMax M3 Thinking
- Claude Sonnet 4.6 Thinking
Intelligence Index
- MiniMax M3 Thinking29.2
- Claude Sonnet 4.6 Thinking30.1
MiniMax M3 Thinking: Better than 70% of 326 scored models
Indexes
Coding Index
Composite of the coding evaluations
- MiniMax M3 Thinking: 58.6
- Claude Sonnet 4.6 Thinking: 63.0Top
Agentic work
AutomationBench-AA
Workflow automation, penalized for guardrail violations
- MiniMax M3 Thinking: 21.3%Top
- Claude Sonnet 4.6 Thinking: 20.1%
AutomationBench-AA tasks completed
Workflows finished end to end without a guardrail violation
- MiniMax M3 Thinking: 4.4%Top
- Claude Sonnet 4.6 Thinking: 3.7%
Harvey LAB-AAv1.1
Legal tasks passed in full, with no material hallucination
- MiniMax M3 Thinking: 0.0%
- Claude Sonnet 4.6 Thinking: —
AA-Briefcasev1.1
Agentic knowledge work, rated head to head
- MiniMax M3 Thinking: 1,089 EloTop
- Claude Sonnet 4.6 Thinking: 1,062 Elo
GDPval-AAv2.1
Economically valuable occupational tasks, rated head to head
- MiniMax M3 Thinking: 1,243 EloTop
- Claude Sonnet 4.6 Thinking: 1,233 Elo
Document reasoning
GDP.pdf
Professional reasoning over PDF documents, all-pass rate
- MiniMax M3 Thinking: 9.8%
- Claude Sonnet 4.6 Thinking: 15.8%Top
AA-LCRv1.1
Reasoning across long documents
- MiniMax M3 Thinking: 83.0%Top
- Claude Sonnet 4.6 Thinking: 80.0%
MLCR-AA
Long-context reasoning over medical records
- MiniMax M3 Thinking: 4.4%
- Claude Sonnet 4.6 Thinking: 7.8%Top
Reasoning and knowledge
Humanity’s Last Exam
Expert-level questions across academic fields
- MiniMax M3 Thinking: 39.0%Top
- Claude Sonnet 4.6 Thinking: 33.6%
GPQA Diamond
Graduate-level science questions
- MiniMax M3 Thinking: 92.9%Top
- Claude Sonnet 4.6 Thinking: 87.5%
IFBench
Following precise output instructions
- MiniMax M3 Thinking: 82.9%Top
- Claude Sonnet 4.6 Thinking: 56.6%
Coding
Terminal-Benchv4.0
Practical software engineering in a terminal
- MiniMax M3 Thinking: 2.0%
- Claude Sonnet 4.6 Thinking: 3.0%Top
SciCode
Scientific computing in Python
- MiniMax M3 Thinking: 47.1%
- Claude Sonnet 4.6 Thinking: 50.1%Top
Legacy evaluations
Terminal-Bench Hard
Earlier terminal task suite
- MiniMax M3 Thinking: 42.4%
- Claude Sonnet 4.6 Thinking: 53.0%Top
τ²-Bench Telecom
Tool use in a customer-support simulation
- MiniMax M3 Thinking: 88.9%Top
- Claude Sonnet 4.6 Thinking: 75.7%
AA-LCR (unversioned)
Long-context reasoning, earlier grading
- MiniMax M3 Thinking: 83.0%Top
- Claude Sonnet 4.6 Thinking: 80.0%
Scores from Artificial Analysis, updated Oct 26, 2026. Artificial Analysis
Live routing data
Performance and reliability
Recent averages across requests routed for this model. Your own speed depends on prompt length, reasoning effort and load.
- Throughput
- 218.4 tok/s
- Faster than 86% of 251 models
- Time to first token
- 7.2 s
- Quicker than 18% of 251 models
- Prompt cache hits
- 81.0%
- About $0.0841 per 1M input tokens with caching
- Request success (24h)
- 99.9%
- Reliable
Cache hits are the share of input tokens recently served from a prompt cache across all requests. They are not a forecast of your own hit rate.
- Healthy (98% or more)
- Degraded (90–98%)
- Down (under 90%)
- No requests
Developers
Use it through the API
Noumen’s API is OpenAI-compatible: any Chat Completions client works with a base URL and a key.
curl https://www.noumen.xyz/api/v1/chat/completions \
-H "Authorization: Bearer $NOUMEN_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "minimax/minimax-m3:thinking",
"messages": [{ "role": "user", "content": "Explain this stack trace." }]
}'import os
from openai import OpenAI
client = OpenAI(base_url="https://www.noumen.xyz/api/v1", api_key=os.environ["NOUMEN_API_KEY"])
reply = client.chat.completions.create(
model="minimax/minimax-m3:thinking",
messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)MiniMax
Related models
Other MiniMax models, starting with the same family.