MiniMax-M3
MiniMax-M3 is served by 2 providers that kRouter can route to, with a 1.0M-token context window and pricing from $0.50 per 1M input tokens. Because more than one provider offers it, the price you pay depends on which one you route to — the same request, the same model, different bills.
Where to get MiniMax-M3
| Provider | Route as |
|---|---|
| Kimchi | kimchi/minimax-m3 |
| CodeBuddy CN | cbcn/minimax-m3 |
What MiniMax-M3 is suited for
A 1.0M window is large enough to hold a substantial part of a codebase in one turn, so it suits whole-repo refactors and long agent sessions where the history keeps growing. It supports extended reasoning, which helps on debugging and multi-step planning but spends output tokens on thinking before the answer appears — worth knowing when you set max_tokens. Tool and function calling are supported, so it can drive an agent loop. Cached input is billed at $0.25 against $0.50 fresh, which is the difference that matters most for agent work — the same context is resent on every turn.
Specs
- Context window
- 1.0M
- Max output
- 512K
- Input / 1M
- $0.50
- Output / 1M
- $2.00
- Extended reasoning
- Tool / function calling
Cheaper alternatives to MiniMax-M3
Comparable capabilities at a lower rate. With a kRouter combo you can send routine turns to one of these and reserve MiniMax-M3 for the work that actually needs it.
- Gemini 2.5 Flash$0.30 in · $2.50 out per 1M
- DeepSeek-V4-Pro$0.43 in · $0.87 out per 1M
- Gemini 2.5 Flash Lite$0.15 in · $1.25 out per 1M
- DeepSeek R1$0.14 in · $0.28 out per 1M
MiniMax-M3 FAQ
How much does MiniMax-M3 cost?
MiniMax-M3 is $0.50 per 1M input tokens and $2.00 per 1M output tokens. Cached input is $0.25, which matters a lot for agent work that resends the same context each turn.
Which providers serve MiniMax-M3?
2 providers: Kimchi, CodeBuddy CN. Through kRouter you can switch between them by changing the route prefix, without changing your tool.
What is MiniMax-M3's context window?
1.0M tokens, with up to 512K tokens of output. kRouter publishes this on /v1/models as context_length, so clients read the real number instead of guessing from the model name.
Is there a cheaper alternative to MiniMax-M3?
Yes — Gemini 2.5 Flash ($0.30/1M in), DeepSeek-V4-Pro ($0.43/1M in), Gemini 2.5 Flash Lite ($0.15/1M in), DeepSeek R1 ($0.14/1M in) are cheaper with comparable capabilities. A kRouter combo can send routine turns to one of those and keep MiniMax-M3 for the work that needs it.
Route to MiniMax-M3 from any tool
kRouter runs locally and gives Claude Code, Cursor, Codex and the rest one endpoint that reaches every provider above — with automatic failover. Free and MIT-licensed.
Install kRouter