Language modelPreview
MiniMax M3.1 Flash Preview
MiniMax
- Released
- -
- Data date
- October 7, 2026
- Input
- Text, images, and video
- Output
- Text
MiniMax M3.1 Flash Preview keeps reasoning on, while letting users choose effort from Low to Max. The provider lists a 1,000,000-token context and text, image, and video input with text output.
Unlike MiniMax M3, reasoning cannot be disabled. MiniMax gives 428 billion total parameters and about 23 billion active parameters for the MoE model.
Specifications and access
| Specification | Value and source |
|---|---|
| Provider model ID | MiniMax-M3.1-Flash-PreviewSource |
| Access | MiniMax Agent Tools and APISource |
| Input modalities | Text, images, and videoSource |
| Output modalities | TextSource |
| Reasoning mode | Always enabled and cannot be turned off; effort can be set from Low to Max.Source |
| Reasoning output | Reasoning content may be returned separately in reasoning_content.Source |
Page 1 of 2
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
ModelDial Backend & testing (low) (Show measurement, test conditions, and source)
- Source value
- 30
- Score
- 30
- Metric
- Backend & testing score out of 100
- Unit
- points
- Category
- private-task-evaluation
- Direction
- Higher is better
- Sample
- 5
- Source type
- independent-evaluator
- Evaluator
- ModelDial
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. ModelDial private tasks for the exact MiniMax-M3.1-Flash-Preview configuration at low reasoning effort. Backend uses five questions worth 20 points each, for 100 points total.
- Context
- Independent evaluator's own private task suite, not a standardized public benchmark. Scores are tied to the displayed reasoning effort and describe those tasks only.
ModelDial Backend & testing (medium) (Show measurement, test conditions, and source)
- Source value
- 34
- Score
- 34
- Metric
- Backend & testing score out of 100
- Unit
- points
- Category
- private-task-evaluation
- Direction
- Higher is better
- Sample
- 5
- Source type
- independent-evaluator
- Evaluator
- ModelDial
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. ModelDial private tasks for the exact MiniMax-M3.1-Flash-Preview configuration at medium reasoning effort. Backend uses five questions worth 20 points each, for 100 points total.
- Context
- Independent evaluator's own private task suite, not a standardized public benchmark. Scores are tied to the displayed reasoning effort and describe those tasks only.
ModelDial Backend & testing (high) (Show measurement, test conditions, and source)
- Source value
- 65
- Score
- 65
- Metric
- Backend & testing score out of 100
- Unit
- points
- Category
- private-task-evaluation
- Direction
- Higher is better
- Sample
- 5
- Source type
- independent-evaluator
- Evaluator
- ModelDial
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. ModelDial private tasks for the exact MiniMax-M3.1-Flash-Preview configuration at high reasoning effort. Backend uses five questions worth 20 points each. Knowledge & reasoning uses 20 cross-disciplinary questions. Each axis is scored out of 100.
- Context
- Independent evaluator's own private task suite, not a standardized public benchmark. Scores are tied to the displayed reasoning effort and describe those tasks only.
ModelDial Knowledge & reasoning (high) (Show measurement, test conditions, and source)
- Source value
- 75
- Score
- 75
- Metric
- Knowledge & reasoning score out of 100
- Unit
- points
- Category
- private-task-evaluation
- Direction
- Higher is better
- Sample
- 20
- Source type
- independent-evaluator
- Evaluator
- ModelDial
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. ModelDial private tasks for the exact MiniMax-M3.1-Flash-Preview configuration at high reasoning effort. Knowledge & reasoning uses 20 cross-disciplinary questions and is scored out of 100.
- Context
- Independent evaluator's own private task suite, not a standardized public benchmark. Scores are tied to the displayed reasoning effort and describe those tasks only.
ModelDial Backend & testing (xhigh) (Show measurement, test conditions, and source)
- Source value
- 50
- Score
- 50
- Metric
- Backend & testing score out of 100
- Unit
- points
- Category
- private-task-evaluation
- Direction
- Higher is better
- Sample
- 5
- Source type
- independent-evaluator
- Evaluator
- ModelDial
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. ModelDial private tasks for the exact MiniMax-M3.1-Flash-Preview configuration at xhigh reasoning effort. Backend uses five questions worth 20 points each, for 100 points total.
- Context
- Independent evaluator's own private task suite, not a standardized public benchmark. Scores are tied to the displayed reasoning effort and describe those tasks only.
ModelDial Frontend & interaction (xhigh) (Show measurement, test conditions, and source)
- Source value
- 15
- Score
- 15
- Metric
- Frontend & interaction score out of 100
- Unit
- points
- Category
- private-task-evaluation
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- ModelDial
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. ModelDial private tasks for the exact MiniMax-M3.1-Flash-Preview configuration at xhigh reasoning effort. Frontend evaluates one web application and its interaction flows out of 100; the method gives no fixed task count.
- Context
- Independent evaluator's own private task suite, not a standardized public benchmark. Scores are tied to the displayed reasoning effort and describe those tasks only.
ModelDial Overall (max) (Show measurement, test conditions, and source)
- Source value
- 69.6
- Score
- 69.6
- Metric
- weighted overall score out of 100
- Unit
- points
- Category
- private-task-evaluation
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- ModelDial
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. ModelDial private tasks for the exact MiniMax-M3.1-Flash-Preview configuration at max reasoning effort. Overall combines Backend 40%, Frontend 30% and Knowledge & reasoning 30%; Backend has five 20-point questions, Frontend evaluates one web application and its flows, and Knowledge & reasoning uses 20 cross-disciplinary questions. Each axis is scored out of 100.
- Context
- Independent evaluator's own private task suite, not a standardized public benchmark. Scores are tied to the displayed reasoning effort and describe those tasks only.
ModelDial Backend & testing (max) (Show measurement, test conditions, and source)
- Source value
- 75
- Score
- 75
- Metric
- Backend & testing score out of 100
- Unit
- points
- Category
- private-task-evaluation
- Direction
- Higher is better
- Sample
- 5
- Source type
- independent-evaluator
- Evaluator
- ModelDial
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. ModelDial private tasks for the exact MiniMax-M3.1-Flash-Preview configuration at max reasoning effort. Backend uses five questions worth 20 points each, for 100 points total.
- Context
- Independent evaluator's own private task suite, not a standardized public benchmark. Scores are tied to the displayed reasoning effort and describe those tasks only.
ModelDial Frontend & interaction (max) (Show measurement, test conditions, and source)
- Source value
- 57
- Score
- 57
- Metric
- Frontend & interaction score out of 100
- Unit
- points
- Category
- private-task-evaluation
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- ModelDial
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. ModelDial private tasks for the exact MiniMax-M3.1-Flash-Preview configuration at max reasoning effort. Frontend evaluates one web application and its interaction flows out of 100; the method gives no fixed task count.
- Context
- Independent evaluator's own private task suite, not a standardized public benchmark. Scores are tied to the displayed reasoning effort and describe those tasks only.
ModelDial Knowledge & reasoning (max) (Show measurement, test conditions, and source)
- Source value
- 75
- Score
- 75
- Metric
- Knowledge & reasoning score out of 100
- Unit
- points
- Category
- private-task-evaluation
- Direction
- Higher is better
- Sample
- 20
- Source type
- independent-evaluator
- Evaluator
- ModelDial
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. ModelDial private tasks for the exact MiniMax-M3.1-Flash-Preview configuration at max reasoning effort. Knowledge & reasoning uses 20 cross-disciplinary questions and is scored out of 100.
- Context
- Independent evaluator's own private task suite, not a standardized public benchmark. Scores are tied to the displayed reasoning effort and describe those tasks only.
ModelDial evaluation time (low) (Show measurement, test conditions, and source)
- Source value
- 201
- Score
- 201
- Metric
- elapsed evaluation time
- Unit
- seconds
- Category
- efficiency
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- ModelDial exact configuration evaluation
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. ModelDial exact MiniMax-M3.1-Flash-Preview configuration at low effort; published evaluation time for backend evaluation. The results snapshot is dated 2026-10-05; the page states the backend was tested on 2026-09-28.
- Context
- This is elapsed time for the listed evaluation suite, not time to first token or one response. Lower efforts have incomplete axis coverage, so times across efforts cover different workloads. Hardware, repeat count and timing boundaries are not stated.
ModelDial evaluation time (medium) (Show measurement, test conditions, and source)
- Source value
- 268
- Score
- 268
- Metric
- elapsed evaluation time
- Unit
- seconds
- Category
- efficiency
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- ModelDial exact configuration evaluation
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. ModelDial exact MiniMax-M3.1-Flash-Preview configuration at medium effort; published evaluation time for backend evaluation. The results snapshot is dated 2026-10-05; the page states the backend was tested on 2026-09-28.
- Context
- This is elapsed time for the listed evaluation suite, not time to first token or one response. Lower efforts have incomplete axis coverage, so times across efforts cover different workloads. Hardware, repeat count and timing boundaries are not stated.
Page 1 of 2
Sources and data date
| Type | Evidence and data date |
|---|---|
| Input | Documented modalities. Retrieved October 7, 2026. |
| Output | Documented modalities. Retrieved October 7, 2026. |
| Research status | Research date October 7, 2026. 4 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not found in the inspected sources | public-standard-benchmarks Original wording MiniMax's official materials publish no numerical standardized benchmark results; ModelDial evaluates its own private task suite. |
| Additional source | ModelDial (retrieved October 7, 2026) |
| Additional source | Model metadata (retrieved October 7, 2026) |
| Additional source | Model metadata (retrieved October 7, 2026) |