impulse

MiniMax M3

Open weights

MiniMax-M3 is a multimodal foundation model from MiniMax.

API id MiniMax-M3Released 31 May 2026 (May 2026)
Input / 1M
$0.30
at MiniMax
Output / 1M
$1.20
at MiniMax
Context window
1M
tokens
Max output
128K
tokens per reply
Hosts
16
18 listings
Released
May 2026
31 May 2026

Pricing

List prices at MiniMax, in US dollars per million tokens.

TierInput / 1MOutput / 1M
Standard$0.30$1.20
Cached input
80% off input
$0.06
Prompts over 512K tokens$0.60$2.40

What it costs a month

At MiniMax list prices, 30 days.

  • Chat assistant5,000 requests a day, 2K in / 0.5K out, cached$158
  • RAG over documents2,000 requests a day, 8K in / 0.6K out, cached$130
  • Coding agent1,000 requests a day, 40K in / 2K out, cached$230
  • Overnight batch20,000 requests a day, 4K in / 0.4K out$1,008
Use your own numbers

Where to run it

16 hosts serve MiniMax M3. The comparison is on blended price, 3 parts input to 1 part output.

HostIn / out per 1M
MiniMaxLab API
MiniMax-M3
$0.30 / $1.20
minimax-m3
$0.225 / $0.90
−25% vs headline
W&B InferenceInference
MiniMax-M3
$0.23 / $0.96
−21% vs headline
GMI CloudInference
MiniMax-M3
$0.24 / $0.96
−20% vs headline
DeepInfraInference
MiniMax-M3
$0.28 / $1.10
−8% vs headline
AIHubMixGateway
minimax-m3
$0.288 / $1.15
−4% vs headline
minimax-m3
$0.30 / $1.20
minimax-m3
$0.30 / $1.20
MiniMax-M3
$0.30 / $1.20
Together AIInference
MiniMax-M3
$0.30 / $1.20
Fireworks AIInference
minimax-m3
$0.30 / $1.20
Novita AIInference
minimax-m3
$0.30 / $1.20
WallyCloud
minimax-m3
$0.33 / $1.32
+10% vs headline
MiniMax-M3
$0.39 / $1.56
+30% vs headline
minimax-m3
$0.396 / $1.58
+32% vs headline
minimax-m3
$0.60 / $2.40
+100% vs headline
Show 2 more listings: regions, snapshots and variants
Fireworks AI
Alias
minimax-m3
$0.30 / $1.20
Inference APIs
Alias
minimax-m3
$0.39 / $1.56
+30% vs headline

Capabilities

Core

  • Streaming
  • Function calling
  • JSON mode
  • Structured outputs

Multimodal

  • Vision
  • Image input
  • Audio
  • Image generation
  • Video

Advanced

  • Reasoning
  • Tool calling
  • MCP
  • Prompt caching
  • Batch API
  • Embeddings
Takes in
TextImages
Produces
Text
Features
Prompt caching
Open weights
License: Apache-2.0
Runs in about 1024.9 GB of GPU memory
Runtimes: SGLang, vLLM

Cheaper alternatives

Current models that cost less and keep most of what this one can do: tools, vision, reasoning, structured output, caching and enough context.

Same model, cheaper host

  • EmpirioLabs AI
    minimax-m3
    $0.225 / $0.90
    25% less
  • W&B Inference
    MiniMax-M3
    $0.23 / $0.96
    21% less
  • GMI Cloud
    MiniMax-M3
    $0.24 / $0.96
    20% less
  • DeepInfra
    MiniMax-M3
    $0.28 / $1.10
    8% less
  • AIHubMix
    minimax-m3
    $0.288 / $1.15
    4% less

Smaller MiniMax models

No cheaper MiniMax model keeps its key capabilities.

Other frontier labs

  • GPT-5.4 Nano
    OpenAI · on OpenAI
    $0.20 / $1.25
    Has every key capability
    12% less
  • Claude Haiku 5.5
    Anthropic · on Anthropic
    $0.10 / $0.50
    Has every key capability
    62% less
  • GPT-6 Luna
    OpenAI · on OpenAI
    $0.10 / $0.50
    Has every key capability
    62% less
  • GPT-4o Mini
    OpenAI · on OpenAI
    $0.15 / $0.60
    Lacks reasoning
    50% less

Open-weight picks

  • Qwen3.7 Plus
    Alibaba Qwen · on Replicate
    $0.276 / $1.10
    Has every key capability
    8% less
  • MiMo V2.6 Pro
    Xiaomi · on GMI Cloud
    $0.348 / $0.696
    Has every key capability
    17% less
  • Muse Glimmer 30B
    Meta · on EmpirioLabs AI
    $0.20 / $0.80
    Has every key capability
    33% less
  • Nemotron 3.5 Super VL
    NVIDIA · on Kyma API
    $0.081 / $0.472
    Has every key capability
    66% less

Model line

Newer and older versions, at each lab's own price.

  1. May 2026MiniMax M3Latest$0.30 / $1.20
  2. Mar 2026MiniMax M2.7$0.21 / $0.84
  3. Feb 2026MiniMax M2.5$0.30 / $1.20
  4. UndatedMiniMax M2.1$0.30 / $1.20
  5. UndatedMiniMax M2$0.30 / $1.20

Call it

Using MiniMax's API.

Python
from litellm import completion

response = completion(
    model="minimax/MiniMax-M3",
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

Quick answers

How much does MiniMax M3 cost?

On MiniMax, MiniMax M3 costs $0.30 per million input tokens and $1.20 per million output tokens. Cached input costs $0.06 per million tokens. The cheapest host is EmpirioLabs AI, at $0.225 per million input tokens and $0.90 per million output tokens.

Which APIs offer MiniMax M3?

MiniMax M3 runs on 16 APIs: MiniMax, EmpirioLabs AI, W&B Inference, GMI Cloud, DeepInfra, AIHubMix, OpenCode Go, Tencent TokenHub, Nebius AI Studio and 7 more.

What is MiniMax M3's context window?

1M tokens, with up to 128K tokens in one reply.

Is MiniMax M3 being retired?

No retirement date is scheduled for it on any API we track.

Is there a newer version of MiniMax M3?

No. MiniMax M3 is the newest model in its line.

What are cheaper alternatives to MiniMax M3?

Models that cost less and keep most of what it can do: GPT-5.4 Nano on OpenAI (12% less), Claude Haiku 5.5 on Anthropic (62% less) and Qwen3.7 Plus on Replicate (8% less). MiniMax M3 itself costs 25% less on EmpirioLabs AI. Savings are on a mix of 3 input tokens to 1 output token.

When was MiniMax M3 released?

MiniMax released it on 31 May 2026.

What is MiniMax M3's API model id?

Use MiniMax-M3 with MiniMax. Other hosts use their own ids: minimax-m3 on EmpirioLabs AI, minimax-m3 on AIHubMix, minimax-m3 on OpenCode Go, minimax-m3 on Tencent TokenHub.

Sources

Raw catalog entry for MiniMax-M3 on MiniMax
{
  "input_cost_per_token": 3e-7,
  "input_cost_per_token_above_512k_tokens": 6e-7,
  "output_cost_per_token": 0.0000012,
  "output_cost_per_token_above_512k_tokens": 0.0000024,
  "cache_read_input_token_cost": 6e-8,
  "cache_read_input_token_cost_above_512k_tokens": 1.2e-7,
  "litellm_provider": "minimax",
  "mode": "chat",
  "supports_function_calling": true,
  "supports_tool_choice": true,
  "supports_prompt_caching": true,
  "supports_reasoning": true,
  "supports_system_messages": true,
  "supports_vision": true,
  "max_input_tokens": 1000000,
  "max_output_tokens": 128000
}