MiniMax M3 Thinking
MiniMax M3 Thinking is the adaptive-thinking version of MiniMax's open-weights frontier model for coding, agent workflows, tool use, long-context tasks, and native multimodal understanding. MiniMax reports 59.0% on SWE-Bench Pro and 66.0% on Terminal Bench 2.1, with Sparse Attention designed to scale context to 1M. It starts with a 512K context cap on NanoGPT for now.
- Reasoning
- Vision
- Tool Calling
- Structured Output
Added Jun 1, 2026
Model weightsPricing
Auto routing · per 1M tokens- Input
- $0.23
- Output
- $0.96
- Cache read
- $0.05
Specifications
- Context window
- 512K
- Max output
- 80K
- Parameters
- 428B / 23B
- Total / active
- Avg output (7d)
- 3.1K tokens
- Longer than 98% of models
Benchmarks
No public benchmark scores for this model yet.
Providers
Choose or restrict providers with the API. Provider-selection prices include the selection markup. Provider selection guide · Pricing API (JSON, USD per 1K tokens)
 | $0.28 | $1.10 | $0.056 | N/A | N/A |
| $0.23 | $0.96 | $0.05 | N/A | N/A |
| $0.242 | $1.01 | $0.0525 | 0.4s | 205 tps | |
| $0.252 | $1.01 | $0.0504 | 2.9s | 55 tps | |
| >512k | $0.504 | $2.02 | $0.101 | ||
| $0.315$0.284 | $1.26$1.13 | $0.063$0.0567 | N/A | N/A | |
| $0.294 | $1.16 | $0.0588 | 1s | 39 tps | |
| $0.315 | $1.26 | $0.063 | 1.6s | 66 tps | |
| $0.315 | $1.36 | $0.0525 | 0.5s | 76 tps | |
| $0.315 | $1.26 | $0.063 | 1.6s | 81 tps | |
| >524k | $0.63 | $2.52 | $0.126 | ||
| $0.315 | $1.26 | $0.063 | 0.6s | 94 tps | |
| $0.347$0.315 | $1.39$1.26 | $0.0693$0.063 | N/A | N/A | |
| $0.315 | $1.26 | $0.063 | 2.2s | 66 tps | |
| $0.315 | $1.26 | $0.063 | 0.8s | 38 tps | |
| $0.63 | $2.52 | $0.126 | N/A | N/A | |
| $0.63 | $2.52 | $0.126 | N/A | N/A |
Privacy:ZDR, prompts not retainedNo training on promptsRetention and training unclearBest latency and throughput are highlighted.
Batch API
Batch rows for this model use normal routing and cost the same as standard API requests. Set provider in a row to choose a provider.