- UBS analysts identified five Chinese AI models released in early 2026 that rival or exceed DeepSeek‘s capabilities, with MiniMax M2.5 as their top pick.
- MiniMax M2.5 delivers performance rivaling Claude Opus 4.6 at roughly one-tenth the price, and its usage has already reached one-third of Anthropic’s Claude volume.
- The other four models include Alibaba’s Qwen 3.5, ByteDance’s Seedance 2.0, and offerings from additional Chinese labs expanding across text, video, and audio generation.
What Happened
UBS analysts published a report in March 2026 highlighting five new Chinese AI models that have moved beyond DeepSeek in capability and market traction. The models span text generation, video creation, and audio synthesis, reflecting the breadth of China’s AI development push over the past year. Rather than a single challenger emerging to replace DeepSeek, the report documents a fragmented but rapidly advancing competitive landscape.
Among the five, UBS singled out MiniMax as their preferred pick. The company, which went public in Hong Kong in January 2026, released its M2.5 model in mid-February with benchmark performance rivaling Anthropic’s Claude Opus 4.6 at a fraction of the cost. UBS analysts said they are positive on MiniMax’s potential upside in gaining share in the global enterprise market.
Why It Matters
DeepSeek’s open-source models dominated headlines throughout 2025, but the Chinese AI landscape has fragmented rapidly. Five separate labs have now shipped models that match or exceed DeepSeek on key benchmarks, suggesting the competitive advantage DeepSeek held has narrowed considerably. For investors and enterprise buyers, this means DeepSeek is no longer the default proxy for Chinese AI capability.
The pricing gap is particularly significant for market dynamics. Data from a mid-February UBS report showed MiniMax’s AI usage had already reached one-third of Anthropic’s Claude volume, at roughly one-tenth the price. For enterprise customers evaluating model providers, cost-per-token differences of that magnitude shift procurement decisions quickly, especially in markets where margins are thin and AI budgets are under scrutiny.
The competitive pressure extends beyond China. MiniMax is explicitly targeting global enterprise customers, putting direct pricing pressure on U.S. providers including OpenAI and Anthropic who charge significantly more per token for comparable benchmark performance.
Technical Details
MiniMax M2.5 leads UBS’s ranking on both price-performance and breadth of capabilities. The model competes across text generation, video generation, ElevenLabs-style audio synthesis, and AI companionship applications. This multimodal breadth distinguishes MiniMax from providers that focus on a single modality. According to data from OpenRouter, developers have been choosing MiniMax M2.5 over DeepSeek’s current V3.2 model and several U.S.-built alternatives for production workloads.
The other models cited by UBS include Alibaba’s Qwen 3.5, a large-scale language model that has consistently ranked among the top performers on coding and reasoning benchmarks, and ByteDance’s Seedance 2.0, which targets the video generation market where OpenAI’s Sora and Google’s Veo have been the primary competitors. Each of the five models targets a different segment of the AI stack, from large-scale language modeling to multimodal content creation.
The diversity of approaches reflects a Chinese AI ecosystem that is no longer converging on a single architecture or company. Multiple well-funded labs are pursuing distinct technical strategies, backed by the large domestic market that provides both training data and initial revenue.
Who’s Affected
Enterprise AI buyers face a broader set of viable Chinese model providers than at any point since the generative AI boom began. The emergence of five competitive options means companies can evaluate providers on price, latency, compliance, and modality coverage rather than defaulting to DeepSeek as the sole Chinese alternative to U.S. models.
For developers building applications on top of third-party models, five competitive Chinese alternatives expand the set of options for cost optimization. This is particularly relevant in price-sensitive markets across Asia, the Middle East, and Latin America where the 10x cost difference between MiniMax and Western providers can determine whether an AI-powered product is economically viable at all.
What’s Next
UBS expects continued model releases from Chinese labs through the rest of 2026, with competition intensifying around multimodal capabilities and enterprise pricing. Whether MiniMax can sustain its usage growth will depend on API reliability, enterprise support infrastructure, and whether its pricing advantage holds as it scales to larger customer volumes. DeepSeek’s V4 release, expected later this year, will test whether the original frontrunner can reclaim its position amid a crowded field.
