On March 18 and 19, two Chinese companies successively released their respective Agent direction large models. The domestic AI startup MiniMax introduced M2.7, while MiMo, under Xiaomi, launched V2-Pro. Both models entered the global first tier in the Agent benchmark, but their API output pricing is 1/21 and 1/8 of Claude Opus 4.6, respectively.

The two companies released their products in the same week, but their offerings are completely different. They represent two distinctly different technology routes, betting on two futures of the Agent era.

Same Exam, 1/17 of Tuition

Let's first look at the most intuitive comparison.

According to data from OpenRouter and the official pricing pages of various companies, the API output price (per million tokens) for MiniMax M2.7 is $1.2, and MiMo-V2-Pro is $3. For reference, Claude Opus 4.6's output price is $25, GPT-5.2 is $14, and Claude Sonnet 4.6 is $15.

The price gap is substantial, but the capability gap is not. In the SWE-bench Verified (the currently most mainstream benchmark for measuring code engineering ability), MiMo-V2-Pro scored 78%, while Sonnet 4.6 scored 79.6%, with a difference of less than two percentage points. M2.7 achieved a SWE-Pro score of 56.22%, which is on par with GPT-5.3-Codex. In VIBE-Pro (end-to-end project delivery capability), M2.7 scored 55.6%, close to Opus 4.6's level.

The key point of this chart is not who is higher or lower—each company's benchmark systems are not entirely aligned, so direct comparisons should be made cautiously. The focus is on the "price-performance gap": domestic Agent models have squeezed into the same capability band but are positioned in completely different price ranges.

Trillion Parameters vs Self-Evolution

Price is just a surface-level indicator. The two companies presented two completely different foundational cards.

MiMo-V2-Pro follows the "big forces create miracles" route. According to Xiaomi's official announcement, V2-Pro has over 1 trillion total parameters, with 42 billion active parameters, supporting an ultra-long context of 1 million tokens. Its core innovation is the Hybrid Attention mechanism, adjusting the ratio of sliding window attention (SWA) to global attention (GA) to 7:1—compared to the previous generation V2-Flash at 5:1. This architecture makes the model more stable when dealing with long documents and parallel calls of multiple tools in Agent scenarios. In PinchBench (Agent tool calling capability assessment), MiMo-V2-Pro achieved a score of 84%.

M2.7 took a completely different path. According to MiniMax's official technical blog released on March 18, the parameter count of M2.7 is not disclosed, but it showcased a mechanism of "self-iterative evolution": the model autonomously runs over 100 optimization cycles, including analyzing failure trajectories, planning modifications, altering its own code architecture, running evaluations, and cycling again, ultimately achieving a 30% performance improvement on the internal evaluation set. In the MLE Bench Lite (machine learning competition difficulty assessment) covering 22 high-difficulty questions, M2.7 won 9 golds, 5 silvers, and 1 bronze, with an average medal rate of 66.6%.

From five dimensions, the two paths have completely different focuses: MiMo-V2-Pro clearly dominates in context length and code engineering, while M2.7 excels in office automation and self-iterative capabilities. According to MiniMax's technical blog, M2.7 scored ELO 1495 in GDPval-AA (office document processing assessment), ranking first among open-source models, and maintained a 97% skill adherence rate in the MM-Claw test, which covers over 40 complex skills.

Four Versions in Five Months

The two companies not only differ in their technological routes but also entirely in their iteration rhythms.

According to public release records, MiniMax iterated four versions from the release of M2 in October 2025 to M2.7 in March 2026, averaging a major version every 49 days. The interval between M2.5 and M2.7 was only about 30 days.

Xiaomi MiMo's rhythm is different: MiMo-7B (an open-source inference model with 7 billion parameters) was released in April 2025, V2-Flash (with 309 billion total parameters) in December of the same year, and V2-Pro (with 1 trillion total parameters) in March 2026. The parameter scales between generations have greater leaps, but the intervals between versions are also longer.

MiniMax chose a strategy of small, rapid steps, with each iteration being small but extremely high in frequency; the self-iterative mechanism of M2.7 is inherently designed for "continuous evolution." Xiaomi chose a strategy of charging up for a powerful strike, with each version representing a significant leap in parameter scale and architecture.

Anonymous for 8 Days, Topped OpenRouter

In addition to technological routes, Xiaomi's release strategy also broke industry conventions.

According to Reuters, on March 11, an anonymous model named Hunter Alpha appeared on the world's largest API aggregation platform, OpenRouter. There was no brand endorsement, no launch event, and no technical blog. Its API pricing was extremely low, but its performance was unexpectedly strong.

The community began to speculate about its origin. According to Republic World and several tech media reports, the most mainstream guess is DeepSeek V4, as MiMo team leader Luo Fuli previously worked on research at DeepSeek. The usage volume skyrocketed, and during the anonymous period, the total number of calls surpassed 1 trillion tokens, topping the weekly chart on OpenRouter.

In the early hours of March 19, Xiaomi revealed: Hunter Alpha is MiMo-V2-Pro. According to the same Reuters report, after the announcement, Xiaomi's Hong Kong stock surged by as much as 5.8%.

This is the first time a domestic large model has proven itself on a global platform through pure blind testing. Not relying on brand, not relying on promotion, it allowed developers to cast their votes with their feet in just 8 days.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。