MiniMax-M3 pricing & specs
MiniMax · released 2026-06-01 · ● verified 2026-10-11
// price history ($/1M tokens)
// cheaper or similar alternatives
| model | in | out | ctx | vs MiniMax-M3 |
|---|---|---|---|---|
| Qwen3.7 Flash | $0.03 | $0.13 | 1M | −89.5% |
| Qwen3.5 Flash | $0.10 | $0.40 | 1M | −66.7% |
| Gemini 2.5 Flash-Lite | $0.10 | $0.40 | 1.05M | −66.7% |
| Qwen3.8 Flash | $0.15 | $0.47 | 1M | −56.2% |
| Qwen3.8 Omni Flash | $0.15 | $0.47 | 1M | −56.2% |
// price by host
| host | in $/1M | out $/1M |
|---|---|---|
| DeepInfra | $0.28 | $1.10 |
| Nebius | $0.30 | $1.20 |
| Novita AI | $0.30 | $1.20 |
// faq
How much does MiniMax-M3 cost?
MiniMax-M3 costs $0.30 per million input tokens and $1.20 per million output tokens. At 50M input and 10M output tokens per month that is about $27.00. Prompts above 512K tokens are billed at $0.60 input and $2.40 output.
What is the context window of MiniMax-M3?
1M tokens, with up to 512K output tokens per request.
Does MiniMax-M3 support prompt caching?
Yes. Cached input tokens cost $0.06 per million.
What are cheaper alternatives to MiniMax-M3?
Qwen3.7 Flash ($0.03 in / $0.13 out), Qwen3.5 Flash ($0.10 in / $0.40 out), Gemini 2.5 Flash-Lite ($0.10 in / $0.40 out) offer the same input types and capabilities for less.
Is MiniMax-M3 open weights?
Yes. It is available from 3 tracked hosts; the cheapest is DeepInfra at $0.28 input and $1.10 output per million tokens.
// spec sheet
- input / 1M
- $0.30
- output / 1M
- $1.20
- cache read / 1M
- $0.06
- > 512K ctx
- $0.60 / $2.40
- blended (3:1)
- $0.525
- context
- 1M
- max output
- 512K
- input types
- text · image · video
- reasoning
- yes
- tool calling
- yes
- structured output
- no
- open weights
- yes
- knowledge cutoff
- n/a
- price source
- MiniMax
// cost calculator
| model | in $/1M | out $/1M | = monthly |
|---|---|---|---|
| Qwen3.7 Flash | $0.03 | $0.13 | $2.80 |
| Gemini 2.5 Flash-Lite | $0.10 | $0.40 | $9.00 |
| Qwen3.5 Flash | $0.10 | $0.40 | $9.00 |
| MiniMax-M3 | $0.30 | $1.20 | $27.00 |
Base-tier list prices. Long-context tiers, caching and batch discounts are not included.