C
ChaoBro
AI Models

MiniMax from M2.7 to M3: The "Office Agent" Breakthrough Route for Chinese Models

by ChaoBro
#MiniMax #M2.7 #M3 #Office Agent #Chinese Models #Model Evolution
MiniMax from M2.7 to M3: The "Office Agent" Breakthrough Route for Chinese Models

Core Conclusion

MiniMax is forging a different product route from DeepSeek, Kimi, and GLM: focusing on Office agent capabilities for office scenarios. M2.7 scored 1514 in GDPval-AA evaluation (ranking fourth), but the upcoming M3 version for the first time demonstrated Office Agent capabilities preview, targeting high-frequency office scenarios such as document processing, PPT generation, and data analysis.

M2.7’s Benchmark Performance

In GDPval-AA (real-world agentic work evaluation), the latest ranking of China’s leading open-source models:

ModelGDPval-AA ScoreMultimodalCodeStrength Scenarios
Xiaomi MiMo-V2.5-Pro1578MediumMedium-HighAgentic workflows
DeepSeek V4 Pro1554HighHighBalanced across scenarios
GLM 5.11535HighHighTool calling + Chinese
MiniMax M2.71514HighMediumMultimodal generation
Kimi K2.61484MediumExtremely HighCode generation

M2.7’s GDPval-AA score is not the highest, but it has a unique advantage: multimodal generation capability is leading among Chinese open-source models. This lays the technical foundation for M3’s Office Agent route.

M3’s New Direction: Office Agent

Based on early May information, the M3 version for the first time demonstrated Office Agent capabilities preview, specifically including:

Office ScenarioCapability DescriptionCompetitor Comparison
PPT GenerationAutomatically generates presentations from topics (content + layout + images)Claude Design can generate design drafts but doesn’t directly output PPT files
Document ProcessingWord document content understanding, format adjustment, summary generationGoogle Gemini has Docs integration but weak Chinese support
Data AnalysisExcel spreadsheet data analysis, chart generation, trend interpretationOpenAI has Code Interpreter but doesn’t support Chinese office scenarios
Meeting AssistantMeeting recording transcription → minutes generation → to-do extractionKimi has long text processing but doesn’t directly generate office documents

Key Differentiation: MiniMax’s Office Agent is not simply “AI + Office”, but embedding multimodal generation capabilities directly into the creation and editing workflow of office documents. Users don’t need to switch tools — AI works right within the document.

May 2026 AI Model Release War

May 2026 is a “super release month” for AI models, with each company’s release plans:

TimeCompanyRelease ContentPositioning
Early MayAnthropicDeveloper Conference + Sonnet 4.8Enterprise AI agents
Mid MayGoogleI/O ‘26 + Gemini 3.2 FlashPersonal AI work layer
MayOpenAIGPT 5.6General capability upgrade
MayMiniMaxM3 + Office AgentOffice scenario agents

MiniMax’s choice to release Office Agent preview at this time point is clear: while giants compete for “general AI” and “developer AI”, MiniMax is seizing the “office AI” niche.

Landscape Assessment

Chinese model competition is shifting from “benchmark score wars” to “scenario differentiation.” Previously, companies were competing on general benchmarks like SWE-bench and MMLU, but now the gap between leading models in general capabilities has narrowed to within 6% (GDPval-AA highest 1578 vs lowest 1484).

The next competitive dimensions will be:

  1. Scenario specialization: Who does better in specific scenarios (office, code, healthcare, legal)
  2. Ecosystem integration: Who better embeds into users’ existing workflows
  3. Cost efficiency: Who offers lower API prices and faster inference at equal capabilities

MiniMax’s Office Agent route is a typical case of the first dimension.

Action Recommendations

RoleRecommendation
Office Scenario UsersIf your core needs are PPT generation, document processing, and data analysis, MiniMax M3 deserves close attention. Do A/B testing after the official release
DevelopersM2.7’s multimodal capabilities are already available via API, suitable for applications needing Chinese + multimodal generation
InvestorsWatch MiniMax’s commercialization path. Office Agent is one of the AI capabilities closest to paid scenarios (enterprise office has strong willingness to pay)

相关内容

GPT-6 Enters Safety Alignment Phase: 5-6 Trillion Parameters, Math Reasoning 92.5%, Code Pass Rate 96.8%

GPT-6 has completed pre-training at the Stargate data center and entered the safety alignment phase. Public data shows math reasoning 92.5%, code generation 96.8%. OpenAI renamed its product department to "AGI Deployment Division", signaling clear all-in commitment.

MiniMax M3 Launching This Month: Targeting Office Scenarios with Major Agentic Capability Upgrades

MiniMax M3 is launching this month, focusing on agentic capability improvements and office scenario adaptation. M2.7 already performed excellently in local model benchmarks, and M3 is expected to further narrow the gap with top-tier models.

GLM-5.1 Lands on 0G Private Computer: What Running a 754B MoE Model Inside a TEE Means

Zhipu AI GLM-5.1, licensed under MIT open-source terms, has launched on 0G Private Computer. The 754B MoE flagship model runs in a Trusted Execution Environment with FP8 quantization, pioneering a new paradigm combining open-source LLMs with privacy computing.