MiniMax M3 vs MiniMax M3.1 Flash
| Benchmark | MiniMax M3 | MiniMax M3.1 Flash |
|---|---|---|
| Humanity's Last Exam | 39 | 48 |
| Terminal-Bench 4.0 | 2 | 25.8 |
| AutomationBench-AA | 21.3 | 72.5 |
| MMMU-Pro (vision) | 78.6 | 79.1 |
Specs
| Spec | MiniMax M3 | MiniMax M3.1 Flash |
|---|---|---|
| Input, per 1M | $0.30 | $0.15 |
| Output, per 1M | $1.20 | $0.60 |
| Cost per task? | $0.51 | $0.22 |
| Speed, tok/s | 109 | 121 |
| Context | 524K | 1M |
| Weights | Open | Closed |
| Released | Jun 2026 | Oct 2026 |
| Memory (4-bit) | 256 GB | — |
| Licence | minimax-community | — |
| Commercial use | Restricted | — |
| See images | ✓ | ✓ |
| Read PDFs | Partial | — |
| Hear audio | — | — |
| Watch video | ✓ | ✓ |
| Call tools | ✓ | ✓ |
| Force a tool call | Partial | ? |
| Return strict JSON | — | — |
| Search the web | ? | ? |
| Run code | ? | ? |
| Use a computer | ? | ? |
What breaks when you switch
MiniMax M3 → MiniMax M3.1 Flash
Complete model response must be appended›
In multi-turn tools, preserve the complete assistant response, including reasoning_content and tool_calls, to maintain reasoning continuity.
MiniMax M3.1 Flash → MiniMax M3
Context: 1M → 524K
Deployment status experimental · Self-hosted›
The documented MiniMax M3 baseline uses an experimental SGLang development image; pin its model revision and image digest and validate your workload.