Language modelActive

MiniMax M3.1 Flash

MiniMax

Released: October 19, 2026
Data date: October 24, 2026

Benchmark-Meta-Score: Unavailable
No independent results
Price: $0.15 · $0.6
Input · Output / 1M tokens
Input: Text, images, and video, with documented tool and thinking inputs.
Output: Text

MiniMax M3.1 Flash has been available since October 19, 2026, through the API, M Plan, and MiniMax Code. Its official weights were publicly accessible on October 24. ModelScope history dates the first commit containing weight files to October 22; the date when they first became public is not established. The model card lists approximately 428 billion total parameters and 23 billion active parameters per token. This multimodal MoE model accepts text, images, and video, with a 1,000,000-token API context window. Its local-weight configuration sets max_position_embeddings to 1,048,576.

The MiniMax Community License permits commercial use under conditions: prominent attribution as “Built with MiniMax M3.1-Flash” is always required. If products or services using the weights generate more than USD 20 million in yearly revenue, prior written authorization is required. Otherwise, a one-time notice to MiniMax is additionally sufficient. This is not an unrestricted open-source license. Reasoning is always on, with effort levels from Low to Max. No unambiguous maximum output length is documented.

Specifications and access

Specification	Value and source
API model ID	MiniMax-M3.1-FlashSource
Availability	Available since October 19, 2026 through the API, M Plan, and MiniMax Code.Source
Reasoning	Always enabled and cannot be disabled. Effort levels are Low, Medium, High, XHigh, and Max; Max is the default.Source
Sampling parameters	Temperature 1.0 and top_p 0.95 in the API documentation.Source
Input modalities	Text, images, and video, with documented tool and thinking inputs.Source
Output modalities	TextSource
Parameter count	The official model card lists approximately 428 billion total parameters and 23 billion active parameters per token.Source
Architecture	MiniMax describes the model as a multimodal MoE. Its published configuration contains 60 language layers, 128 routed experts, 4 selected experts per token, one shared expert, and 32 vision layers.Source
Weight configuration position limit	The local-weight configuration sets max_position_embeddings to 1,048,576. The API documentation lists a 1,000,000-token context window, so the configuration field is reported separately from the published API limit.Source
Weight availability	The official weights were publicly accessible and ungated on October 24. ModelScope history dates the first commit with weight files to October 22, 2026 at 14:37:30 UTC; when public access began is not established. The October 17 Hugging Face repository creation timestamp does not prove earlier availability.Source
Weight license	MiniMax Community License. Commercial use for products or services requires prominent attribution as “Built with MiniMax M3.1-Flash”. If those products or services generate more than USD 20 million in yearly revenue, prior written authorization is required. Otherwise, only a one-time notice to MiniMax is additionally required.Source
Quantization	The configuration specifies dynamic MXFP8 and NVFP4 for MoE experts. The license applies to this official weight release; third-party quantizations may differ.Source
Direct MiniMax API maximum output	Not stated unambiguously for the direct MiniMax API. Published OpenRouter output limits apply to the specific provider endpoint listed there.Source
Release	2026-10-19Source
Model identity	released M3.1 Flash, distinct from M3.1 Flash PreviewSource
Context window	1,000,000 tokensSource
Maximum output	not stated unambiguously for the direct MiniMax APISource
Input modalities	text, image, videoSource
Output modalities	textSource
Reasoning modes	Low, Medium, High, XHigh, MaxSource
Sampling parameters	temperature 1.0; top_p 0.95Source
API prices in USD per million tokens	USD 0.15 input; USD 0.60 output; USD 0.015 cache read per million tokensSource
Priority tier	unsupportedSource
Cache write price	not listedSource
Output speed	170+ tokens/s P50; daily peak 200+ tokens/sSource
Agent benchmark cost	USD 111 total for Agents’ Last ExamSource
Total parameters in billions	Approximately 428 billion total parameters according to the official model cardSource
Active parameters in billions	Approximately 23 billion active parameters per token according to the official model cardSource
Architecture and expert routing	MiniMaxM3SparseForConditionalGeneration; 60 text layers; 128 routed experts; 4 routed experts per token; 1 shared expert; 32 vision layers; max_position_embeddings 1,048,576; dynamic MXFP8 and NVFP4 MoE quantizationSource
Local weights position limit	max_position_embeddings: 1,048,576 in the official local-weight configurationSource
First visible official weight-shard commit	First commit containing weight files: October 22, 2026, 14:37:30 UTCSource
Public weights availability	The official weights were publicly accessible and ungated on October 24, 2026.Source
Commercial use license conditions	MiniMax Community License: commercial use for products or services requires prominent attribution as “Built with MiniMax M3.1-Flash”. If those products or services generate more than USD 20 million in yearly revenue, prior written authorization is required. Otherwise, only a one-time notice to MiniMax is additionally required.Source
Provider model ID	OpenRouter model ID: minimax/minimax-m3.1-flash Hugging Face bridge ID listed by OpenRouter catalog: MiniMaxAI/MiniMax-M3.1-FlashSource
Deployment options	Relace [relace/fp4]; endpoint name=Relace \| minimax/minimax-m3.1-flash-20261019; quantization=fp4; raw status code=0 InferenceNet [inference-net/fp4]; endpoint name=InferenceNet \| minimax/minimax-m3.1-flash-20261019; quantization=fp4; raw status code=0 InferenceNet [inference-net/fp4/us]; endpoint name=InferenceNet \| minimax/minimax-m3.1-flash-20261019; quantization=fp4; raw status code=0 GMICloud [gmicloud/fp8]; endpoint name=GMICloud \| minimax/minimax-m3.1-flash-20261019; quantization=fp8; raw status code=0 DigitalOcean [digitalocean]; endpoint name=DigitalOcean \| minimax/minimax-m3.1-flash-20261019; quantization=unknown; raw status code=0 Novita [novita/fp16]; endpoint name=Novita \| minimax/minimax-m3.1-flash-20261019; quantization=fp16; raw status code=0 Nebius [nebius/fp4]; endpoint name=Nebius \| minimax/minimax-m3.1-flash-20261019; quantization=fp4; raw status code=0 Wafer [wafer]; endpoint name=Wafer \| minimax/minimax-m3.1-flash-20261019; quantization=unknown; raw status code=0 ModelRun [modelrun/fp4]; endpoint name=ModelRun \| minimax/minimax-m3.1-flash-20261019; quantization=fp4; raw status code=0 Parasail [parasail/fp8]; endpoint name=Parasail \| minimax/minimax-m3.1-flash-20261019; quantization=fp8; raw status code=0 Together [together]; endpoint name=Together \| minimax/minimax-m3.1-flash-20261019; quantization=unknown; raw status code=0 Minimax [minimax]; endpoint name=Minimax \| minimax/minimax-m3.1-flash-20261019; quantization=unknown; raw status code=0 SambaNova [sambanova]; endpoint name=SambaNova \| minimax/minimax-m3.1-flash-20261019; quantization=unknown; raw status code=0 (This lists every provider row for the exact OpenRouter route(s) queried. Status codes are preserved as returned, not interpreted or used to filter offers.)Source
Context values by source	Relace [relace/fp4]; context_length=1048576 tokens InferenceNet [inference-net/fp4]; context_length=1000000 tokens InferenceNet [inference-net/fp4/us]; context_length=1000000 tokens GMICloud [gmicloud/fp8]; context_length=1048576 tokens DigitalOcean [digitalocean]; context_length=1048576 tokens Novita [novita/fp16]; context_length=1000000 tokens Nebius [nebius/fp4]; context_length=1048576 tokens Wafer [wafer]; context_length=1048576 tokens ModelRun [modelrun/fp4]; context_length=1048576 tokens Parasail [parasail/fp8]; context_length=1048576 tokens Together [together]; context_length=1048576 tokens Minimax [minimax]; context_length=1048576 tokens SambaNova [sambanova]; context_length=1048576 tokens (These are limits per OpenRouter provider route. They do not replace separately documented direct-API or model-configuration values.)Source
Source output limit	Relace [relace/fp4]; max_completion_tokens=943718 tokens InferenceNet [inference-net/fp4]; max_completion_tokens=524288 tokens InferenceNet [inference-net/fp4/us]; max_completion_tokens=524288 tokens GMICloud [gmicloud/fp8]; max_completion_tokens=943718 tokens DigitalOcean [digitalocean]; max_completion_tokens=943718 tokens Novita [novita/fp16]; max_completion_tokens=524288 tokens Nebius [nebius/fp4]; max_completion_tokens=943718 tokens Wafer [wafer]; max_completion_tokens=943718 tokens ModelRun [modelrun/fp4]; max_completion_tokens=943718 tokens Parasail [parasail/fp8]; max_completion_tokens=524288 tokens Together [together]; max_completion_tokens=943718 tokens Minimax [minimax]; max_completion_tokens=524288 tokens SambaNova [sambanova]; max_completion_tokens=943718 tokens (These are output limits for the corresponding OpenRouter provider route, not direct-API limits. Provider differences remain separate.)Source
API pricing in USD per million tokens	Relace [relace/fp4]; prompt USD/1M=0.09; completion USD/1M=0.5; cache-read USD/1M=0.01; raw discount field=0 InferenceNet [inference-net/fp4]; prompt USD/1M=0.1; completion USD/1M=0.55; cache-read USD/1M=0.01; raw discount field=0 InferenceNet [inference-net/fp4/us]; prompt USD/1M=0.1; completion USD/1M=0.55; cache-read USD/1M=0.01; raw discount field=0 GMICloud [gmicloud/fp8]; prompt USD/1M=0.1425; completion USD/1M=0.57; cache-read USD/1M=0.01425; raw discount field=0.05 DigitalOcean [digitalocean]; prompt USD/1M=0.15; completion USD/1M=0.6; cache-read USD/1M=0.015; raw discount field=0 Novita [novita/fp16]; prompt USD/1M=0.15; completion USD/1M=0.6; cache-read USD/1M=0.015; raw discount field=0 Nebius [nebius/fp4]; prompt USD/1M=0.15; completion USD/1M=0.6; cache-read USD/1M=0.015; raw discount field=0 Wafer [wafer]; prompt USD/1M=0.15; completion USD/1M=0.6; cache-read USD/1M=0.015; raw discount field=0 ModelRun [modelrun/fp4]; prompt USD/1M=0.15; completion USD/1M=0.6; cache-read USD/1M=0.015; raw discount field=0 Parasail [parasail/fp8]; prompt USD/1M=0.15; completion USD/1M=0.6; cache-read USD/1M=0.015; raw discount field=0 Together [together]; prompt USD/1M=0.15; completion USD/1M=0.6; cache-read USD/1M=0.015; raw discount field=0 Minimax [minimax]; prompt USD/1M=0.15; completion USD/1M=0.6; cache-read USD/1M=0.015; raw discount field=0 SambaNova [sambanova]; prompt USD/1M=0.3; completion USD/1M=1.2; cache-read USD/1M=0.03; raw discount field=0 (Per-token API prices were multiplied exactly by 1,000,000. Raw discount fields and returned pricing schedules are preserved separately. No endpoint pricing object in these returned rows contained an input_cache_write/cache-write price; it is not reported as zero.)Source
API capabilities	Relace [relace/fp4]; supported_parameters=["reasoning","include_reasoning","temperature","top_p","top_k","min_p","stop","max_tokens","seed","logit_bias","frequency_penalty","presence_penalty","repetition_penalty","response_format","tools","tool_choice","reasoning_effort"]; supports_tool_choice={"auto":true,"function":false,"none":true,"required":false}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false InferenceNet [inference-net/fp4]; supported_parameters=["reasoning","include_reasoning","max_tokens","temperature","top_p","stop","frequency_penalty","presence_penalty","seed","top_k","min_p","logit_bias","response_format","structured_outputs","repetition_penalty","tools","tool_choice","reasoning_effort"]; supports_tool_choice={"auto":true,"function":true,"none":true,"required":true}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false InferenceNet [inference-net/fp4/us]; supported_parameters=["reasoning","include_reasoning","max_tokens","temperature","top_p","stop","frequency_penalty","presence_penalty","seed","top_k","min_p","logit_bias","response_format","structured_outputs","repetition_penalty","tools","tool_choice","reasoning_effort"]; supports_tool_choice={"auto":true,"function":true,"none":true,"required":true}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false GMICloud [gmicloud/fp8]; supported_parameters=["reasoning","include_reasoning","max_tokens","temperature","top_p","seed","response_format","tools","tool_choice","reasoning_effort"]; supports_tool_choice={"auto":true,"function":false,"none":true,"required":false}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false DigitalOcean [digitalocean]; supported_parameters=["reasoning","include_reasoning","temperature","top_p","max_tokens","response_format","structured_outputs","tools","tool_choice","logprobs","top_logprobs","reasoning_effort"]; supports_tool_choice={"auto":true,"function":true,"none":true,"required":true}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false Novita [novita/fp16]; supported_parameters=["reasoning","include_reasoning","max_tokens","temperature","top_p","stop","frequency_penalty","presence_penalty","seed","top_k","repetition_penalty","response_format","tools","tool_choice","reasoning_effort"]; supports_tool_choice={"auto":true,"function":false,"none":true,"required":false}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false Nebius [nebius/fp4]; supported_parameters=["reasoning","include_reasoning","max_tokens","temperature","top_p","stop","frequency_penalty","presence_penalty","seed","top_k","logit_bias","logprobs","top_logprobs","response_format","structured_outputs","tools","tool_choice","reasoning_effort"]; supports_tool_choice={"auto":true,"function":true,"none":true,"required":true}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false Wafer [wafer]; supported_parameters=["reasoning","include_reasoning","temperature","top_p","top_k","min_p","frequency_penalty","presence_penalty","repetition_penalty","stop","seed","max_tokens","logit_bias","response_format","structured_outputs","tools","tool_choice","reasoning_effort"]; supports_tool_choice={"auto":true,"function":true,"none":true,"required":true}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false ModelRun [modelrun/fp4]; supported_parameters=["reasoning","include_reasoning","max_tokens","temperature","presence_penalty","repetition_penalty","frequency_penalty","stop","top_p","structured_outputs","tools","tool_choice","reasoning_effort"]; supports_tool_choice={"auto":true,"function":true,"none":true,"required":true}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false Parasail [parasail/fp8]; supported_parameters=["reasoning","include_reasoning","max_tokens","temperature","top_p","frequency_penalty","presence_penalty","repetition_penalty","seed","stop","top_k","logit_bias","response_format","structured_outputs","tools","tool_choice","reasoning_effort"]; supports_tool_choice={"auto":true,"function":true,"none":true,"required":true}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false Together [together]; supported_parameters=["reasoning","include_reasoning","max_tokens","temperature","top_p","stop","frequency_penalty","presence_penalty","top_k","repetition_penalty","logit_bias","min_p","response_format","structured_outputs","tools","tool_choice","logprobs","top_logprobs","reasoning_effort"]; supports_tool_choice={"auto":true,"function":true,"none":true,"required":true}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false Minimax [minimax]; supported_parameters=["reasoning","include_reasoning","max_tokens","temperature","top_p","tools","tool_choice","response_format","reasoning_effort"]; supports_tool_choice={"auto":true,"function":false,"none":false,"required":false}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false SambaNova [sambanova]; supported_parameters=["reasoning","include_reasoning","max_tokens","temperature","top_p","top_k","stop","response_format","structured_outputs","tools","tool_choice","reasoning_effort"]; supports_tool_choice={"auto":true,"function":true,"none":true,"required":true}; supports_implicit_caching=true; native_tools={}; supports_voice_cloning=false; supports_multiple_audio_references=false; supports_image_reference=false; supports_diarization=false (Supported parameters and returned API capability flags are listed per provider route. The route-level architecture returned in the same response is: {"input_modalities":["text","image","video"],"instruct_type":null,"modality":"text+image+video->text","output_modalities":["text"],"tokenizer":"Other"}.)Source
Quantized variant	Relace [relace/fp4]; tag=relace/fp4; quantization=fp4 InferenceNet [inference-net/fp4]; tag=inference-net/fp4; quantization=fp4 InferenceNet [inference-net/fp4/us]; tag=inference-net/fp4/us; quantization=fp4 GMICloud [gmicloud/fp8]; tag=gmicloud/fp8; quantization=fp8 DigitalOcean [digitalocean]; tag=digitalocean; quantization=unknown Novita [novita/fp16]; tag=novita/fp16; quantization=fp16 Nebius [nebius/fp4]; tag=nebius/fp4; quantization=fp4 Wafer [wafer]; tag=wafer; quantization=unknown ModelRun [modelrun/fp4]; tag=modelrun/fp4; quantization=fp4 Parasail [parasail/fp8]; tag=parasail/fp8; quantization=fp8 Together [together]; tag=together; quantization=unknown Minimax [minimax]; tag=minimax; quantization=unknown SambaNova [sambanova]; tag=sambanova; quantization=unknown (Tags and quantization values are taken literally from endpoint responses. Missing values are not inferred from names.)Source

Benchmark meta-analysis

Overall and area scores

These Gradually indexes place selected published measurements on a 0 to 100 scale. They are not accuracy percentages and do not guarantee results in your application.

Overall index

Not enough independent results for a score yet

provisional

Test series: 0
Test families: 0 / 65
Sources: 0
Share of the fixed test panel (informative): 0 %

Documented method · Included test families and measurements

Method version: benchmark-meta-score-v3

The index includes active measurements from reviewed primary sources when at least five models are compared in the same documented comparison slice across dataset, harness, agent, tools, and trial count. Each benchmark series and evaluator contributes one shared setup slice from the newest capture. When one source and series include several explicitly documented reasoning tiers, the highest documented tier (none through max) is selected, never the best result value. The explicit source label “Non-reasoning” is treated as none. Unknown or ambiguous tiers exclude the model from that series. Within each series, percentages and rates are logit-transformed, other units stay unchanged, and all values are then standardized. A one-factor model then estimates one shared strength per model and, per series, how sharply it separates the models. Area scores move away from that strength only as far as the model's own results in the area support. The overall score weights the core areas at 75% and finance, law and health at 25%. It requires at least eight test series from two sources and three areas. The 0 to 100 scale refers to a fixed reference group of 29 language models whose average sits at 50.

Published model-specific parameters can differ. This is not a controlled experiment; comparisons with such differences are marked provisional.

Unmeasured series are not entered as zero. Where a model has no own results in a core area, its overall strength stands in, and the note “estimated without own results in …” names the area. Fewer test series lead to a larger standard error.

The 95% interval follows from the standard error of 200 recalculations with randomly redrawn test families. It describes how precisely the recorded test series determine the value. It is no prediction for future tasks or for later recalculations with other participants.

The source count is not a count of independent experiments.

OECD handbook on composite indicators

All 0 test families and 0 of 0 raw measurements are listed here. The complete calculation for every model is available in the JSON export.

Benchmark sources at a glance

Each evaluator measures different capabilities under its own conditions. Test series, composite indices and their components stay separate. The detailed results below retain their scales, configurations and source dates.

Evaluator	Test series	Composite indices	Recorded active measurements
Artificial Analysis	14	7	185
Vals AI	10	1	146
LMArena	0	0	4

Coverage is not a quality ranking. Overlapping tests and multiple settings do not count as independent experiments. Vendor claims, historical and disputed results are excluded from this overview and remain labelled in the detailed evidence.

163 measurements from these sources still lack a verified test-family identity and do not increase the test-series count.

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.

Automation-Bench v1.0.6 (Show measurement, test conditions, and source)

Source value: 62.8
Score: 62.8
Metric: public tasks
Unit: %
Category: agent/work
Direction: Higher is better
Sample: 600
Source type: vendor-reported
Evaluator: MiniMax, M3.1 Flash benchmark table
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Claude Code harness; max effort.

Source MiniMax, M3.1 Flash benchmark table

BankerToolBench (Show measurement, test conditions, and source)

Source value: 76.2
Score: 76.2
Metric: score
Unit: %
Category: agent/work
Direction: Higher is better
Source type: vendor-reported
Evaluator: MiniMax, M3.1 Flash benchmark table
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. OpenCode 1.17.18; MiniMax M2.7 as judge.
Context: MiniMax does not report a model-specific sample count here.

Source MiniMax, M3.1 Flash benchmark table

Terminal-Bench 4.0 (Show measurement, test conditions, and source)

Source value: 38.1
Score: 38.1
Metric: score
Unit: %
Category: coding
Direction: Higher is better
Source type: vendor-reported
Evaluator: MiniMax, M3.1 Flash benchmark table
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Codex 0.149.1.
Context: MiniMax does not report a model-specific sample count here.

Source MiniMax, M3.1 Flash benchmark table

LOCA-Bench 256K (Show measurement, test conditions, and source)

Source value: 60.3
Score: 60.3
Metric: score
Unit: %
Category: long-context
Direction: Higher is better
Source type: vendor-reported
Evaluator: MiniMax, M3.1 Flash benchmark table
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Official implementation; 256K context setting.
Context: MiniMax does not report a model-specific sample count here.

Source MiniMax, M3.1 Flash benchmark table

Agents’ Last Exam (Show measurement, test conditions, and source)

Source value: 29.6
Score: 29.6
Metric: official Overall setting
Unit: %
Category: agent/work
Direction: Higher is better
Sample: 152
Source type: vendor-reported
Evaluator: MiniMax, M3.1 Flash benchmark table
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Official ALE Overall setting; Claude Code; max effort.

Source MiniMax, M3.1 Flash benchmark table

OSWorld 2.0 (Show measurement, test conditions, and source)

Source value: 61.7
Score: 61.7
Metric: score
Unit: %
Category: computer use
Direction: Higher is better
Source type: vendor-reported
Evaluator: MiniMax, M3.1 Flash benchmark table
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Official recommended 0808 full release.
Context: MiniMax does not report a model-specific sample count here.

Source MiniMax, M3.1 Flash benchmark table

PostTrain Bench v1.1 (Show measurement, test conditions, and source)

Source value: 39.3
Score: 39.3
Metric: score
Unit: %
Category: coding
Direction: Higher is better
Source type: vendor-reported
Evaluator: MiniMax, M3.1 Flash benchmark table
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Claude Code 2.1.251; maximum output 512K.
Context: MiniMax does not report a model-specific sample count here.

Source MiniMax, M3.1 Flash benchmark table

DeepSWE v1.1 (Show measurement, test conditions, and source)

Source value: 73.7
Score: 73.7
Metric: score
Unit: %
Category: coding
Direction: Higher is better
Source type: vendor-reported
Evaluator: MiniMax, M3.1 Flash benchmark table
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. mini-SWE-agent 2.4.5; max effort; 512K output cap.
Context: MiniMax does not report a model-specific sample count here.

Source MiniMax, M3.1 Flash benchmark table

NL2Repo-Bench (Show measurement, test conditions, and source)

Source value: 61.2
Score: 61.2
Metric: score
Unit: %
Category: coding
Direction: Higher is better
Source type: vendor-reported
Evaluator: MiniMax, M3.1 Flash benchmark table
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Claude Code 2.1.233; max effort; 512K output cap; two inconsistent spec/test tasks corrected; reward-hacking score zero.
Context: MiniMax does not report a model-specific sample count here.

Source MiniMax, M3.1 Flash benchmark table

SWE Refactor Bench (Show measurement, test conditions, and source)

Source value: 23
Score: 23
Metric: score
Unit: %
Category: coding
Direction: Higher is better
Source type: vendor-reported
Evaluator: MiniMax, M3.1 Flash benchmark table
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Claude Code 2.1.220; max effort; 200K output cap.
Context: MiniMax does not report a model-specific sample count here.

Source MiniMax, M3.1 Flash benchmark table

CyberGym (Show measurement, test conditions, and source)

Source value: 82.2
Score: 82.2
Metric: score
Unit: %
Category: cyber security
Direction: Higher is better
Source type: vendor-reported
Evaluator: MiniMax, M3.1 Flash benchmark table
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Claude Code 2.1.263; max effort; 1M output cap.
Context: MiniMax does not report a model-specific sample count here.

Source MiniMax, M3.1 Flash benchmark table

BabyVision w/tools (Show measurement, test conditions, and source)

Source value: 91
Score: 91
Metric: score
Unit: %
Category: multimodal
Direction: Higher is better
Source type: vendor-reported
Evaluator: MiniMax, M3.1 Flash benchmark table
Status: active
Retrieved at: 2026-10-24
Methodology: Vendor-reported result. Claude Code 2.1.285; image minimum side 1,500 pixels.
Context: MiniMax does not report a model-specific sample count here.

Source MiniMax, M3.1 Flash benchmark table

Page 1 of 36

Sources and data date

Type	Evidence and data date
Price	Base API prices without caching or batch discounts. Higher context tiers and other rates are listed under Costs. Source: Pricing · MiniMax official pricing (retrieved October 24, 2026)
Input	Documented modalities. Retrieved October 24, 2026. Source: MiniMax Anthropic-compatible API reference · MiniMax Anthropic-compatible API reference (retrieved October 24, 2026) · MiniMax Anthropic-compatible API reference · Editorial description reviewed October 24, 2026
Output	Documented modalities. Retrieved October 24, 2026. Source: MiniMax Anthropic-compatible API reference · MiniMax Anthropic-compatible API reference (retrieved October 24, 2026) · MiniMax Anthropic-compatible API reference · Editorial description reviewed October 24, 2026
Research status	Research date October 24, 2026. 166 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Not found in the inspected sources	Direct MiniMax API maximum output Note The checked direct MiniMax API documentation did not unambiguously state a maximum output-token count. OpenRouter endpoint-specific limits are documented separately.
Not published	Cache write price Note The official pricing table lists cache-read cost but no cache-write rate for this model.
Not published	Knowledge cutoff Note The exact model card, pinned configuration, release post, and official API documents do not state a training-data or knowledge cutoff.
Unresolved	First public weight availability date Note The exact official weights are public and ungated on October 24, but neither the release post nor a dated access record establishes when public access first began. The October 22 weight commit and October 17 Hugging Face createdAt timestamp do not prove public access on those dates.
Unresolved	Current OpenCompass results Note The current public OpenCompass results could not be fully retrieved during the source check. Possible results for this exact model therefore remain unresolved.
Not found in the inspected sources	Exact-model ARC Prize results Note The full current public ARC Prize evaluations and model registry were read; no exact identity for this target appears across ARC-AGI-1/2/3. Similar names without an exact model identity bridge were not mapped.
Not found in the inspected sources	Exact-model Scale Labs results Note The Scale Labs index and current public result boards in the bounded capture set were read; no exact row for this target was found. Kimi-K3 and undated DeepSeek V4.1 Flash rows are not bridged to newer targets.
Additional source	MiniMax M3.1 Flash release announcement (retrieved October 24, 2026) · MiniMax M3.1 Flash release announcement · Editorial description reviewed October 24, 2026
Additional source	MiniMax model guide (retrieved October 24, 2026) · MiniMax model guide · Editorial description reviewed October 24, 2026
Additional source	MiniMax M3.1 Flash official weight repository (retrieved October 24, 2026)
Additional source	MiniMax official ModelScope weight repository history (retrieved October 24, 2026)
Additional source	MiniMax Community License (retrieved October 24, 2026) · MiniMax M3.1 Flash Community License at pinned Hugging Face revision · Editorial description reviewed October 24, 2026
Additional source	MiniMax M3.1 Flash model configuration (retrieved October 24, 2026) · MiniMax M3.1 Flash configuration at pinned Hugging Face revision · Editorial description reviewed October 24, 2026
Additional source	MiniMax M3.1 Flash DSpark configuration (retrieved October 24, 2026)
Additional source	Artificial Analysis MiniMax M3.1 Flash model page (retrieved October 24, 2026)
Additional source	DuelLab, GameBench 2 model result (retrieved October 24, 2026)
Additional source	LiveBench (retrieved October 24, 2026)
Additional source	LiveBench (retrieved October 24, 2026)
Additional source	LiveBench (retrieved October 24, 2026)
Additional source	LMArena (retrieved October 24, 2026)
Additional source	Vals AI (retrieved October 24, 2026)
Additional source	Vals AI (retrieved October 24, 2026)
Additional source	Vals AI (retrieved October 24, 2026)
Additional source	Vals AI (retrieved October 24, 2026)
Additional source	Vals AI (retrieved October 24, 2026)
Additional source	Vals AI (retrieved October 24, 2026)
Additional source	Vals AI (retrieved October 24, 2026)
Additional source	Vals AI (retrieved October 24, 2026)
Additional source	Vals AI (retrieved October 24, 2026)
Additional source	Vals AI (retrieved October 24, 2026)
Additional source	Vals AI (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026) · MiniMax M3.1 Flash model card at pinned Hugging Face revision · Editorial description reviewed October 24, 2026
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	Model metadata (retrieved October 24, 2026)
Additional source	MiniMax M3.1 Flash official ModelScope weight repository · Editorial description reviewed October 24, 2026

Specifications and access

Overall and area scores

Overall index

Benchmark sources at a glance

Measurements without matching peer values

Published prices

Sources and data date