Benchmark Almanac

MiniMax

We track 27 benchmarks that MiniMax made or helped make. Its models are on its lab page.

4 in use 2 with public code 7 with a leaderboard here

Benchmarks

Name Kind Released Best score Status
VIBE-Pro (MiniMax) Agents build full-stack projects for Web, Android, iOS and Windows, including server APIs and databases. Claude Code based verifiers check interaction logic and visual output. Building whole projects Feb 2026 55.6% Fading
MM-ClawBench Agents complete common OpenClaw tasks such as study planning, office documents, scheduled research, and code maintenance. Scoring is accuracy. General assistants Mar 2026 75.4% Fading
RISE Agents answer real questions from human experts that need multi-step search and interaction with information-dense web pages. MiniMax has not published the scoring method. Deep research Feb 2026 64.3% Fading
MiniMax TTS Multilingual Test Set Text-to-speech models clone two Common Voice speakers per language and read 100 test sentences in each of 24 languages. Scored by word error rate and speaker similarity. Speech recognition and translation May 2025 0.36% In use
Finance Modeling Pro Agents complete financial modeling tasks in Excel. Expert rubrics score each task over three trials. Finance May 2026 57% System cards only
HyperTask Agents build applications over long vibe coding sessions with about 100 feature or step requirements per task. Expert rubrics score the result over three trials. Building whole projects May 2026 67.6% System cards only
MEWC v2 Models solve 100 hard Microsoft Excel World Championship problems with Excel tools. Expert rubrics score the answers over three trials. Documents and spreadsheets May 2026 63.3% System cards only
GDPval-MM Agents produce office deliverables for GDPval tasks in a MiniMax agent framework. A judge model compares full trajectories pairwise and reports a win rate. Tasks from real jobs Feb 2026 73.5% Fading
OctoCodingbench Agents complete 72 coding tasks while following rules from system prompts, AGENTS.md files, skills, and memory. An LLM judge checks 2,422 pass or fail items. Fixing real repositories Dec 2025 36.2% Fading
SWE-Review Models review code changes in several languages and scenarios to find defects. Scores reflect defect recall and the rate of hallucinated defects. Fixing real repositories Dec 2025 16.2% Fading
VIBE (MiniMax) Agent builds complete web, simulation, Android, iOS or backend apps from scratch. Agent verifiers check execution, interaction and visuals. Building whole projects Dec 2025 90.7% Fading
Coding (MiniMax in-house) Models answer coding requests from real Hailuo AI users. The report does not describe the scoring. Function-level problems Jan 2025 90.2% System cards only
Creative Writing (MiniMax in-house) Models write creative pieces for real Hailuo AI user requests. Automatic assessments score the writing. Creative writing Jan 2025 81.3% System cards only
Finance Modeling Agents complete expert-written financial modeling problems with Excel tools, from research to analysis. Expert-designed rubrics score each problem, averaged over three runs. Finance Feb 2026 21.6% Fading
GDPval-Rubrics (MiniMax) MiniMax rubric grading of model deliverables on public GDPval tasks, in an environment matched to GDPval-AA. Tasks from real jobs Jun 2026 74.8% In use
General Assistant (MiniMax in-house) Models answer real assistant queries from Hailuo AI users. Automatic assessments score the replies. Overall chat quality Jan 2025 73.9% System cards only
Hard Capability (MiniMax in-house) Models handle complex productivity requests with several levels of instructions that leading models often fail. Following precise instructions Jan 2025 64.8% System cards only
Instruction Following (MiniMax in-house) Models follow multi-level instructions collected from user cases where current models fail. Following precise instructions Jan 2025 46.3% System cards only
Knowledge Q&A (MiniMax in-house) Models answer knowledge questions drawn from real Hailuo AI user queries. Trivia and open questions Jan 2025 78.6% System cards only
Long Context (MiniMax in-house) Models translate, summarize and analyze long documents from real user requests. Reasoning over long documents Jan 2025 93.8% System cards only
MEWC Models solve 179 Microsoft Excel World Championship problems from 2021 to 2026 with Excel tools. Output cells are compared one by one with answer cells. Documents and spreadsheets Feb 2026 74.4% Fading
Safety (MiniMax in-house) Models respond to safety-relevant user prompts. MiniMax scores whether replies are safe and helpful. Refusing harmful requests Jan 2025 90.9% System cards only
SVG-Bench (MiniMax) Model creates or edits SVG graphics from text or image prompts. A VLM checks results, averaged over 3 runs. Web and app front ends Jun 2026 63.7% In use
VIBE-V2 (MiniMax) Agents build front-end and full-stack Web, Android and iOS projects from scratch in Claude Code. An agent-as-verifier pipeline checks interaction logic and visual output. Building whole projects Jun 2026 50.1% In use
MR-NIAH (Multi-Round Needles-In-A-Haystack) Models must repeat one earlier answer from a long multi-turn dialogue history of 2K to 1M tokens. Scored by adjusted recall of key parts. Finding facts in long inputs Jan 2025 System cards only
Role-play Benchmark (MiniMax) Role-play agents are tested on situated reenactment in Chinese and English scenes. Grading targets the immersive experience rather than a verifiable answer. Creative writing Jan 2026 84.7 Not in cards
SynLogic Models solve synthetic reasoning problems, including Sudoku, arithmetic puzzles, and mazes. The port uses official task graders on separate easy and hard validation sets. Language reasoning May 2025 Not in cards