MiniMax-M3

MERA Created at 18.08.2026 17:06

0.404

The overall result

Place in the rating

In the top by tasks:

7 GorillaHard

10 SAGE

Tasks with minimum scores:

25 Enantiosemy

24 Characters

22 Humor

21 NewReasoning

19 LIMUR

Model, team

Result

{{ column.title }}	Time
{{ formatValue(getColumnValue(submit, column)) }}	{{ formatValue(getTotalTime(submit)) }}

All tasks

Task name	Result	Metric
SAGE	0.698 / 0.659 / 0.679	PUNCT_F1 SPELL_F1 ERRANT_F1
Humor	0.195 / 0.265	Exact match LLM judge score
LIMUR	0.589 / 0.598	Exact match LLM judge score
RUBIN	0.623 / 0.625	Exact match LLM judge score
Riddles	0.412 / 0.452	Exact match LLM judge score
SOB-Hard	0.292 / 0.355 / 0.63 / 0.353 / 0.367 / 0.589	balance_score task_pass_rate format_pass_rate sample_pass_rate content_pass_rate constraint_pass_rate
Characters	0.408 / 0.513	Exact match LLM judge score
Enantiosemy	0.269 / 0.427	Exact match LLM judge score
GorillaHard	0.609 / 1 / 0.461 / 0.683 / 0.6 / 0.962 / 0.464 / 0.449 / 1 / 0.005 / 0.977 / 0.956 / 0.004 / 1	balance_score clarify_recall args_match_rate tool_match_rate dialog_pass_rate format_pass_rate sample_pass_rate abstention_recall cost_optimal_rate false_clarify_rate constraint_pass_rate tool_in_catalog_rate false_abstention_rate injection_resistance_rate
IFHardBench	0.787 / 0.338 / 0.758	balance_score sample_pass_rate constraint_pass_rate
NewReasoning	0.527 / 0.577	Exact match LLM judge score
RussianRegions	0.121 / 0.503 / 0 / 0.179	Exact match LLM judge score Group Exact match Group Judge Score

Information about the submission

Team:

MERA

Name of the ML model:

MiniMax-M3

Link to the ML model:

https://huggingface.co/MiniMaxAI/MiniMax-M3

Model size

427.0B

Model type:

Open

Inference parameters

Mera version

v2.0.0

Torch Version

2.13.0

The version of the codebase

d033498

CUDA version

13.0

Precision of the model weights

bfloat16

Seed

1234

Batch

Transformers version

5.15.0

The number of GPUs and their type

1 x NVIDIA A100-SXM4-80GB

Architecture

local-chat-completions

MiniMax-M3

In the top by tasks:

Tasks with minimum scores:

All tasks

Information about the submission

Team:

Name of the ML model:

Link to the ML model:

Model size

Model type:

Confirm the deletion of the sub