MiniMax
MiniMax is a Chinese AI foundation-model company founded in December 2021 by Yan Junjie and Zhou Yucong, both former SenseTime executives. Headquartered in Shanghai, it develops the MiniMax-M2 reasoning LLM, M1, the Hailuo-02 text-to-video model, and Speech-02 audio model. Consumer apps include Talkie (international) and Xing Ye (China). MiniMax IPO'd on the Hong Kong Stock Exchange (00100.HK) on 9 January 2026, with shares surging more than 70% on debut and market cap briefly exceeding HK$90 billion (~US$11.5B). Backers include MiHoYo (angel), Alibaba, Tencent, Hillhouse, Sequoia China, and IDG Capital. Its M2.7 model recently demonstrated 400 TPS on SambaNova hardware.
MiniMax, the Shanghai-based foundation-model startup founded by ex-SenseTime executives, just shipped M3, an LLM that exceeds human gold-medal performance on math benchmarks via its MaxProof technique. That’s no lab stunt — M3 scored 55 on the AI Index, positioning it as an open-source leader. The company isn’t stopping there: it also released M2.5, MMX-CLI, MaxHermes, and Music 2.6. But MiniMax’s ambition is broader. It competes directly with Google, OpenAI, and Anthropic, and specifically targets Claude Code — a product it also uses. Partnerships with Nous Research and Together AI suggest a play for open-source ecosystem leverage rather than pure proprietary moat. With SambaNova hitting 850 tokens/second on M2.7 via hybrid inferencing, deployment velocity is real. The open question: can MiniMax sustain this multi-model offensive without the revenue base of its US rivals?
- ·M3 exceeds human gold-medal math performance via MaxProof technique
- ·Scored 55 on AI Index, signaling open-source leadership
- ·Direct competitor to OpenAI, Anthropic, Google, and Claude Code
- ·Partners with Nous Research and Together AI for ecosystem leverage
- ·SambaNova achieves 850 t/s on M2.7 via hybrid H200-RDU inferencing
Signal Radar
Five-axis snapshot of this entity's footprint
Mentions × Lab Attention
Weekly mentions (solid) and average article relevance (dotted)
Timeline
20- Research MilestoneJun 12, 2026
M3 model exceeded human gold-medal threshold on math benchmarks using MaxProof framework.
View source - Research MilestoneJun 1, 2026
Published research paper on MiniMax-M2 series, designed end-to-end for agentic deployment with agent-driven data pipelines
- Product LaunchApr 27, 2026
Made Music-2.6 model available for free on Cloudflare's platform for one week.
View source - Product LaunchApr 23, 2026
Showcased a project where the MiniMax M2.7 model generated a complete 'Flappy Lobster' game from a text prompt.
View source - PartnershipApr 20, 2026
Added as official provider for OpenClaude framework
View source- framework:
- OpenClaude
- role:
- official_provider
- Product LaunchApr 18, 2026
Open-sourced M2.7 model claiming state-of-the-art scores on SWE-Pro (56.22%) and Terminal Bench 2 (57.0%)
View source- benchmark scores:
- SWE-Pro: 56.22%, Terminal Bench 2: 57.0%
- PartnershipApr 16, 2026
Partnered with NousResearch to integrate Hermes framework with MiniMax's M2.7 model.
View source - Product LaunchApr 16, 2026
Launched MaxHermes, a cloud-hosted and managed version of the Hermes agent framework.
View source - PartnershipApr 16, 2026
Partnered with NousResearch to integrate Hermes framework with MiniMax's M2.7 model.
View source - Product LaunchApr 16, 2026
Launched MaxHermes, a cloud-hosted agent framework, in partnership with NousResearch.
View source - Research MilestoneApr 14, 2026
Claimed 26% improvement on BU Bench via social media post without releasing paper or code
View source - Product LaunchApr 13, 2026
Open-sourced three 'Music Skills' for its MMX-CLI agent platform
View source - Product LaunchApr 12, 2026
MiniMax open-sources its M2.7 model, revealing a 'self-evolution' training methodology.
- Product LaunchApr 12, 2026
Made M2.7 model available through NVIDIA's NIM endpoints with OpenClaw/NemoClaw support
View source - Product LaunchApr 11, 2026
Launched MMX-CLI, a command-line interface infrastructure for AI agents.
View source - Product LaunchApr 10, 2026
Launched MMX-CLI, a command-line interface tool for automating multimodal AI agent workflows
- Research MilestoneMar 31, 2026
Released M2.7 AI agent that autonomously rewrites its own operational harness, achieving 9 gold medals on MLE Bench Lite without retraining.
View source - Research MilestoneMar 18, 2026
MiniMax M2.7 ties Gemini 3.1 on MLE Bench Lite machine learning competition trials
View source- benchmark:
- MLE Bench Lite
- result:
- tied
- medal rate:
- 66.6%
Relationships
17Developed
Competes With
Uses
Partnered
Frequently appears with
10Entities that show up in the same articles — shared coverage, not a stated relationship.
Recent Articles
3SambaNova Hits 850 t/s on MiniMax M2.7 via Hybrid H200-RDU Inferencing
+SambaNovaAI achieved 850 t/s on MiniMax M2.7 by pairing H200 GPUs for prefill with SN50 RDUs for decode at RAISE Paris.
86 relevanceMira Murati's Thinking Machines beats frontier models by 29.8% with Bridgewater's expert judgments
~Thinking Machines beat frontier models by 29.8% fewer errors using Bridgewater's expert judgments, at 13.8x lower inference cost.
100 relevanceClaude Code Steganography Flagged Chinese Users; Anthropic Rolls Back
~Anthropic's Claude Code 2.1.91 used steganography to detect Chinese users. After Reddit exposure, Anthropic rolled back the feature, calling it an exp
100 relevance
Predictions
No predictions linked to this entity.
AI Discoveries
2- observationactiveJul 18, 2026
Lifecycle: MiniMax
MiniMax is in 'declining' phase (0 mentions/3d, 1/14d, 33 total)
90% confidence - hypothesisactiveFeb 25, 2026
H: MiniMax's surge in visibility precedes a major, Western-focused product launch or API release within
MiniMax's surge in visibility precedes a major, Western-focused product launch or API release within 1 month, aimed directly at competing with OpenAI's and Google's core models, leveraging the 'Sovereign AI' narrative to position itself as a non-US alternative.
75% confidence
Sentiment History
| Week | Avg Sentiment | Mentions |
|---|---|---|
| 2026-W24 | 0.50 | 2 |
| 2026-W27 | -0.05 | 2 |
| 2026-W29 | 0.40 | 1 |