Loading page
MiniMax
MiniMax-M3
MiniMax-M3 is a chat-oriented large language model from MiniMax, listed here with measured serving throughput of roughly 107 output tokens per second.
Input from
$0.200 / 1M tokens
across 7 providers
API Pricing
Cheapest on EcoHash — 33% below avg| Provider | Input / 1M | Output / 1M | Cached / 1M |
|---|---|---|---|
| $0.200 | $0.900 | - | |
| $0.280 | $1.10 | $0.056 | |
| $0.300 | $1.20 | - | |
| $0.300 | $1.20 | - | |
| $0.300 | $1.20 | $0.060 | |
| $0.300 | $1.20 | $0.060 | |
| $0.300 | $1.20 | $0.060 | |
| $0.400 | $2.00 | $0.100 |
Prices updated daily. Last check: Sep 8, 2026
MiniMax-M3 pricing by provider
Input, output, and batch rates, plus alternatives, for one provider at a time.
Performance & Benchmarks
Source: Artificial Analysis →Intelligence
35.7 / 100
Coding
58.6 / 100
Output Speed
95.8 t/s
Latency (TTFT)
1.0s
Reasoning & Knowledge
- GPQA Diamond92.9%
- Humanity's Last Exam39.0%
Coding
- SciCode47.1%
Agentic & Tool Use
- Terminal-Bench Hard42.4%
- Terminal-Bench v2.165.2%
- τ²-bench88.9%
- τ-bench Banking15.3%
Instruction & Long Context
- IFBench82.9%
- Long-Context Reasoning83.0%
Benchmarks measured Sep 2026. Scores are independent evaluations, not vendor-reported.
Model Details
General
- Creator
- MiniMax
- Modalities
- Text
Capabilities
- Tool Calling
- No
- Open Source
- No
Strengths & Limitations
Strengths
- Measured output throughput of about 106.8 tokens per second, suitable for streaming responses in interactive chat interfaces
- Time to first token of roughly 936 ms, keeping perceived latency under one second in the measured configuration
- Positioned as a general-purpose chat model, so it can be evaluated as a drop-in for instruction-following and assistant workloads
- Performance figures sourced from a third-party benchmarker (Artificial Analysis) rather than vendor-reported numbers
- Part of MiniMax's ongoing series of general-purpose language models, giving it a documented lineage of prior releases to compare against
- Listed across the providers shown in the pricing table on this page, allowing direct cost and availability comparison
Limitations
- We do not track a confirmed context window for MiniMax-M3, so long-document suitability must be verified with your provider
- No task benchmark scores (coding, reasoning, math, multilingual) are recorded in our database for this model
- Confirmed modality support is not tracked, so image or audio input should not be assumed
- Throughput and latency figures reflect a single measured serving configuration and will vary by provider, region, and load
- Provider availability for MiniMax models is generally narrower than for the most widely hosted open-weight families
Key Features
•Chat and instruction-following model interface
•Measured generation throughput of ~106.8 output tokens/second
•Measured time to first token of ~936 ms
•Streaming-friendly latency profile for interactive assistants
•Third-party performance measurement via Artificial Analysis
•Multi-provider price comparison in the table on this page
•Part of the MiniMax model series
About MiniMax-M3
MiniMax-M3 is a text chat model developed by MiniMax, a Chinese AI company that has released several generations of general-purpose language models under the MiniMax name. In our catalog it is classified as a chat model, meaning it is intended for conversational and instruction-following workloads served through an inference API rather than for a narrow single task such as embedding or transcription.
The measurements we track for MiniMax-M3 come from Artificial Analysis: approximately 106.8 output tokens per second of generation throughput and about 936 ms time to first token. Those two numbers describe interactive responsiveness — how quickly a first token appears after a request, and how fast the remainder of the response streams. A sub-second time to first token and triple-digit token-per-second output place it in a range that is generally workable for streaming chat interfaces and assistant-style products, though both figures depend on the provider, region, hardware, and load at the time of measurement.
We do not currently track a confirmed context window, modality list, or task benchmark scores for MiniMax-M3 in our database, so this page does not make claims about those. If you are evaluating the model, check the documentation of the specific provider you plan to use for context length, supported input types, tool calling support, and any published evaluation results, and use the pricing table on this page to compare what each provider charges for serving it.
Common Use Cases
MiniMax-M3 is a general-purpose chat model, so the natural fit is conversational workloads where response latency matters: customer-facing assistants, in-product chat, drafting and rewriting tools, summarization of pasted text, and back-end steps in an application pipeline that need natural-language output. The measured sub-second time to first token and roughly 107 tokens per second of generation make it a candidate for streaming UIs where users watch text appear rather than waiting for a complete response. Because we do not have confirmed context window, modality, or task-benchmark data for this model, workloads that depend on very long inputs, image or audio understanding, or verified performance on coding and math evaluations should be validated with your own test set and with the provider's documentation before committing.
Frequently Asked Questions
How much does MiniMax-M3 cost?
Pricing varies by provider and by pricing type — input tokens, output tokens, and any cached or batch rates are billed differently, and providers change rates over time. Check the pricing table on this page for current per-provider rates rather than relying on a fixed figure.
What is MiniMax-M3 best used for?
It is catalogued as a chat model, so it suits conversational assistants, instruction-following tasks, drafting, rewriting, and summarization. Its measured latency profile — about 936 ms to first token and roughly 107 output tokens per second — makes it reasonable for streaming, user-facing interfaces.
How fast is MiniMax-M3?
In measurements from Artificial Analysis, MiniMax-M3 produced about 106.8 output tokens per second with a time to first token of roughly 936 ms. Both numbers depend on the serving provider, hardware, region, and concurrent load, so treat them as a reference point rather than a guarantee.
What context window does MiniMax-M3 support?
We do not have a confirmed context window for MiniMax-M3 in our database. Because context limits can also differ between hosts of the same model, verify the maximum input length with the specific provider you plan to use.
Does MiniMax-M3 accept image input or support tool calling?
We do not track confirmed modality or tool-calling details for this model, so we make no claim either way. Consult the API documentation of the provider serving MiniMax-M3 for the exact input types and function-calling features it exposes.
Who created MiniMax-M3?
MiniMax-M3 was developed by MiniMax, which has released multiple generations of general-purpose language models under the MiniMax name. This entry covers the M3 chat model specifically.