MiniMax
12 published articles
AI Video Generation
MiniMax H3 prices video at a third of rivals, ranks No. 1 in editing
MiniMax H3 packs text, image, video and audio into one model, prices output at a third of mainstream per-second rates and ranks first in video editing on Artificial Analysis. The weights are promised within days. How the output holds up outside the company's own demos is the open question.
2026-08-06
Analysis
Benchmarks, not bargains: China's AI labs reach parity
Chinese AI labs Qwen, DeepSeek, MiniMax, and Kimi have matched or beaten US frontier models in reasoning, coding, and document parsing benchmarks. With aggressive pricing and open-weight strategies, they are no longer just cheap alternatives, they are real competitors.
2026-07-27
AI Research
MiniMax's VTP learned to understand before it learned to generate
MiniMax's VTP framework rethinks how visual tokenizers are pre-trained, replacing pure reconstruction objectives with a mix that rewards semantic understanding. The result: a tokenizer that actually scales with compute, and a 65.8% FID improvement on downstream generation.
2026-07-22
Artificial Intelligence
The reward-hacking collapse that nearly killed MiniMax's proof model
MiniMax details how M3's proof capabilities survived a reward-hacking crisis that nearly killed the project. The four-layer verifier and MaxProof test-time framework pushed scores above human gold-medal thresholds on IMO 2025 and USAMO 2026, offering a blueprint for any lab dealing with adversarial model behavior.
2026-07-16
LLM Performance
A Chinese video-generation startup just quietly beat Claude Opus at coding
MiniMax's M2.7 scores 56.22% on SWE-Pro, matching near-Claude Opus performance, while touting 97% skill adherence on complex tasks and superior office productivity editing. The model signals a shift from benchmark chasing to real-world agent deployment.
2026-07-14
Artificial Intelligence
The 'um' that makes AI sound human is finally here
MiniMax's Speech 2.8 adds native filler words, breathing, and hesitation to AI speech, solving the 'too perfect' problem that made synthetic voices robotic. The model also delivers 10-second voice cloning and cross-language accuracy.
2026-07-11
AI Labs & Research
MiniMax's M3 just wrote its own CUDA kernel, and opened the code
MiniMax M3 scores 83.5 on BrowseComp, edges past Opus 4.7, and handles up to 1M tokens natively. In a remarkable autonomy test, it self-optimized a GPU kernel from 7.6% to 71.3% peak utilization without human intervention.
2026-07-09
AI Model Release
MiniMax launches M2.7 model with strong software engineering and office productivity skills
MiniMax's M2.7 model delivers strong results in software engineering benchmarks and professional office tasks, with a 97% skill adherence rate on complex instructions and an ELO score of 1495 on GDPval-AA.
2026-07-08
Artificial Intelligence
MiniMax just shipped a model for every AI job you can name
Chinese AI startup MiniMax launches M3, Hailuo 2.3, MiniMax Code, and new speech/music models, broadening its product lineup in a competitive landscape.
2026-07-06
Artificial Intelligence
MiniMax's M3 just beat Opus 4.7 at browsing, trained itself, and never asked for help
MiniMax M3 delivers a 9.4x CUDA kernel speedup, beats Opus 4.7 on BrowseComp, and autonomously replicated an ICLR paper. All in an open-weight package, and it never asked for help.
2026-07-06
Artificial Intelligence
MiniMax's new M2.5 coding model tops the benchmark at 5% of the price
MiniMax's M2.5 model tops the Multi-SWE-Bench coding benchmark, beats mainstream models on workspace tasks, and costs a tenth to a twentieth of competitors. Open-source weights are on HuggingFace.
2026-07-04
AI Research & Development
Ma Jiaqi taught MiniMax engineers a hard lesson about forgotten tokens
MiniMax's internal investigation into why its M2 model couldn't output the name 'Ma Jiaqi' revealed a structural mismatch between pre-training vocabulary and post-training data distribution. The root cause: low-frequency tokens' lm_head vectors drift during SFT, losing generation ability while retaining understanding. A full-vocabulary coverage fix resolved the issue and also mitigated language mixing in Japanese.
2026-07-04