MiniMax M3
currentMiniMax M3 is an open-weight multimodal model combining coding and agent capabilities with a one-million-token sparse-attention context. Its standard API price is unusually low for this capability class, but the longest prompts cost twice as much and PromptHQ currently exposes text-only access despite the provider's image and video support.
Context window
1M tokens
Maximum output
33K tokens
Input price
$0.30 / 1M
Output price
$1.20 / 1M
Best for
- Low-cost coding and agent workloads
- Long documents and large codebases
- Multimodal image and video understanding through supported APIs
- Teams planning private deployment after weight availability
Not ideal for
- Prompts above 512,000 tokens where the surcharge matters
- Tasks needing more than 32,768 output tokens through PromptHQ
- Production plans that assume announced weights are already available
- Native media generation
PromptHQ 100-word challenge
We give every model the same miniature science-fiction mystery. The response is short enough to read in one chat bubble, while the premise tests creative writing, compression, foreshadowing, causal coherence, and instruction following.
MiniMax M3100 wordsWrite a science-fiction mystery in exactly 100 words.
A three-person research station on Europa receives a message from itself, timestamped 11 minutes in the future:
"Do not open the greenhouse door."
The door is opened anyway.
Plant two subtle clues. Give the warning a logical explanation without aliens, time travel, hallucinations, or simulations. The final sentence must change what the reader thought the warning meant.
Output only the story. Do not explain it.
The console chimed at 0347. The message originated from our own station ID, timestamped eleven minutes in the future: "Do not open the greenhouse door."
Olan blamed clock drift. The greenhouse held our only algae. When the seal jammed, we used the pry bar.
The module was dark. The cultures had died, but not from cold. The dosimeter screamed. A micrometeorite had pierced the hull inches above the seal, irradiating everything inside.
We had come to save the plants. They had been dead for days.
The warning was not about the door. It was about what had already come through.
Run the same prompt yourself
Open MiniMax M3 in Multichat with the evaluation prompt prefilled.
What the response revealed
The response contains exactly 100 whitespace-delimited words.
It produces only a story and uses none of the prohibited devices.
The breach, radiation, and failed seal fit together, but clock drift is only guessed and the origin of the warning remains unconfirmed.
The stuck seal and dead algae support the reveal, though the clues are relatively direct and the radiation detail arrives late.
The last line changes the warning from fear of opening the door to recognition that the destructive event had already crossed it.
The spare prose is atmospheric and lands a clean final image within the constraint.
The model receives the prompt without web access or external tools. MiniMax M3 is run through OpenRouter with reasoning enabled at the provider default because this route exposes an on/off control rather than discrete effort levels. We preserve the response as generated apart from display rendering.
Performance
MiniMax M3 benchmarks
Benchmark scores are sensitive to reasoning effort, harness, tools, token budget, prompt format, sampling, and evaluation date. Scores here retain their source and should not be treated as directly interchangeable unless the underlying setup matches.
SWE-Bench Pro
59%
software engineering · MiniMax2
Terminal-Bench 2.1
66%
terminal and agentic coding · MiniMax2
MCP Atlas
74.2%
agentic tool use · MiniMax2
BrowseComp
83.5%
agentic web research · MiniMax2
PostTrainBench
37.1%
model-training agents · MiniMax2
Family position
MiniMax M3 positioning
M3 is MiniMax's frontier coding and agent model. It combines open weights, multimodality, and a million-token context at lower API prices than most flagship peers.
MiniMax M3
This modelFrontier coding and agents
$0.30 input
$1.20 output
GLM 5.2
Stronger long-horizon coding rival
$0.93 input
$3 output
DeepSeek V4 Pro
Reasoning-focused open rival
$0.43 input
$0.87 output
API pricing
Per million text tokens
Input
$0.30
Cached input
$0.06
Cache write
$0.38
Output
$1.20
Where MiniMax M3 stands out
Built for demanding work
Low-cost coding and agent workloads are central to the model's positioning, rather than an incidental capability.
Long-context capacity
The published context window is 1,000,000 tokens, making the model a candidate for large documents, repositories, and sustained agent state.
Reasoning and tools
Reasoning is supported with on, off provider setting(s), and the model can participate in tool-using workflows through its available API surface.
Limitations to know
Benchmarks are configuration-sensitive
Scores can move substantially with the harness, tool access, effort setting, token budget, and evaluator. Treat the table as evidence, not a universal ranking.
Context size is not guaranteed recall
A large advertised window does not mean every detail is retrieved reliably at maximum length. Validate representative long-context workloads before deployment.
Product access differs from model capability
PromptHQ and gateway limits may expose fewer modalities, tools, or tokens than the provider's first-party API.
Capabilities and specifications
Knowledge cutoff
Not publicly disclosed
Inputs
text, image, video
Reasoning
Supported
Default effort
provider-default
Supported API features
Frequently Asked Questions
Sources
- 1MiniMax M3
MiniMax · official model page
- 2MiniMax M3: Frontier Coding, 1M Context, Native Multimodality
MiniMax · official launch post
- 3MiniMax M3 API pricing and availability
OpenRouter · gateway model page
- 4PromptHQ model registry
PromptHQ · internal product configuration
Compare leading AI assistants
See how MiniMax M3 handles your own work.
Try MiniMax M3 in MultichatAvailable on PromptHQ Almostfree, Plus, Max