The frontier US labs might still have a lead in AI, but China is nipping at their heels.
China-based MiniMax’s M2 model has placed fifth on the Analysis Intelligence Index. The model beat out prominent models including Google’s current flagship model 2.5 Pro and Claude 4.`1 Opus. It also became the highest-placed Chinese model on the list, beating Alibaba’s Qwen3 235B and DeepSeek V3.2.
“We’re open-sourcing MiniMax M2 — Agent & Code Native, at 8% Claude Sonnet price, ~2x faster,” the company said on X. It added that MiniMax M2 had advanced coding capabilities, and was engineered for end-to-end developer workflows. It also claimed that MiniMax M2 had strong agentic performance with robust handling of long-horizon toolchains including mcp, shell, browser, retrieval, code. MiniMax also said their model was “Smarter, Faster, Cheaper” with efficient parameter activation.
MiniMax had some impressive benchmark results to back up those claims. On most popular benchmarks, it was comfortably ahead of Gemini 2.5 Pro, and beat Anthropic Sonnet 4.5 on some benchmarks as well. It also beat GPT-5 on the FinSearchComp-global benchmark.
MiniMax is yet another Chinese lab that has been producing some very capable models without a lot of fanfare. The company was founded in Shanghai, China in December 2021 by former engineers from SenseTime. It has quickly risen to prominence as one of China’s top “AI Tiger” companies, receiving significant investment from major players such as Alibaba, Tencent, Hillhouse, and MiHoYo. MiniMax has also confidentially filed for a Hong Kong IPO, targeting a valuation above $4 billion.
MiniMax produces a range of models. The MiniMax M1 & M2 are foundation models using hybrid Mixture-of-Experts architectures and proprietary attention mechanisms, capable of processing up to one million text tokens at once. The Hailuo 02 is a state-of-the-art image-to-video generation achieving top leaderboard placements in global technical benchmarks. In addition, MiniMax also has a speech-to-text model, a MiniMax agent for dynamic knowledge tasks, and Talkie, and AI avatar for education and interactive digital storytelling.
And MiniMax isn’t the only Chinese company that’s looking to compete with — and outshine — the top US labs. Alibaba, DeepSeek, Moonshot AI and many others find a place on the top leaderboards. Unlike US models, these models are generally open-source, and can often be much cheaper than the US counterparts. The US does seem to have a lead at the moment, with with so many Chinese companies producing high-quality models, it’s not a given that this lead will last in the coming years.