Spirit AI's Spirit v1.6 Tops Authoritative Benchmarks, Closes 1.5 Billion Yuan Series A+ Round | Oasis Capital Vitality

Spirit AI, an Oasis Capital seed portfolio company, has scored twin breakthroughs in technology and capital.

The company's self-developed embodied foundation model, Spirit v1.6, ranked #1 globally on the RoboArena leaderboard, becoming the first Chinese team to top this authoritative international benchmark. Simultaneously, it closed a 1.5 billion yuan Series A+ round, bringing total funding to nearly 5 billion yuan within three months.

In the AI era, China has remained a vital participant in global innovation. And in embodied intelligence — a critical frontier — Chinese teams are now stepping to the front rank of technical competition. Spirit v1.6's ascension to the top of RoboArena marks not only the first time a Chinese embodied model has taken the lead in a mainstream international evaluation system, but also demonstrates the continued breakthroughs Chinese teams are making in next-generation AI core technologies.

Today, Spirit AI officially announced two major milestones:

The company's self-developed embodied foundation model Spirit v1.6 topped RoboArena — known as the "Olympics" of North American embodied intelligence — outperforming NVIDIA's Cosmos3 and Physical Intelligence's Pi0.5, breaking Silicon Valley's lock on the leaderboard to become the first Chinese embodied model to reach #1.

Full results: https://robo-arena.github.io/leaderboard

At the same time, Spirit AI secured another 1.5 billion yuan in Series A+ funding, with investors including top-tier dollar funds, major strategic investors, and state-backed funds. Existing shareholders doubled down, assembling a premium capital lineup spanning top financial PE firms, international dollar funds, industrial strategic capital, and local government investment platforms.

At this critical inflection point for the industry, Spirit AI has achieved a "three-pronged" advance: technical breakthroughs, capital backing, and commercial deployment. With generational leaps in proprietary technology, the company is systematically raising the bar for domestic embodied intelligence and anchoring a robust "China technology force" in the global map of physical general intelligence.

The "Olympics" of North American Embodied Intelligence: The RoboArena Leaderboard

RoboArena is an open international embodied intelligence robot strategy evaluation platform jointly initiated by UC Berkeley, Stanford University, NVIDIA, and other top global academic institutions and tech giants. It has been called the "Chatbot Arena" of the robotics field. Breaking from traditional single-lab assessment models, it is widely recognized as a "world-class authoritative master ranking" with substantial credibility and lasting industry relevance — the "Olympics" of embodied intelligence.

RoboArena's fairness and authority rest on several core dimensions:

• Open international and distributed collaboration: Evaluations rely on a standardized robot hardware network deployed across dozens of top universities and research institutions worldwide. Global volunteer evaluators conduct cross-border, cross-institutional open testing in diverse real-world physical environments, ensuring broad technical applicability.
• Double-blind pairwise comparison: Pioneering A/B double-blind testing in robotics, evaluators — completely unaware of model developers' identities — conduct cross-blind assessments of different strategies executing identical tasks, eliminating subjective bias entirely.
• Dynamic Elo rating algorithm: Abandoning easily gamed static benchmarks, it adopts the classic Elo dynamic rating system from competitive sports. Through hundreds of international real-world physical matchups, it objectively and authentically reflects each model strategy's core generalization capabilities in complex real-world environments.
• Fully open-source and tamper-proof: The evaluation framework, data pipeline, and ranking algorithm are fully open-sourced to the international community. The process is completely transparent, auditable, and reproducible, establishing industry consensus on its neutrality and authority.

In the embodied intelligence era, RoboArena marks the transition of robotics from "laboratory self-certification" to "cross-international large-sample, real-world-based open competition." Ranking among the top on this leaderboard signifies that an algorithmic model has reached world-class levels in multi-task performance, strong generalization, and industrial-grade deployment.

Model Tops the Charts: Spirit v1.6 Outperforms NVIDIA Cosmos3 and PI0.5

The core driver of continued capital commitment is Spirit AI's hard technical breakthroughs in embodied foundation models. At a moment when global embodied intelligence competition is converging on model generalization and real-world physical operation capabilities, the company's self-developed embodied foundation model Spirit v1.6 achieved the #1 comprehensive score on the authoritative RoboArena embodied intelligence benchmark, surpassing NVIDIA Cosmos3 and Physical Intelligence PI0.5 — another instance of a Chinese embodied model overtaking international industry benchmarks.

Across multiple real-world scenario double-blind A/B tests on the platform, Spirit v1.6 demonstrated all-around generalization and operational advantages. Facing complex scenarios including flip-lid opening and closing, fine-grained object grasping, and multi-step continuous tasks, the model formed complete, fluid action loops, performing robustly in environmental understanding, target recognition, motion planning, and execution — fully validating Spirit AI's extreme engineering and algorithmic capabilities.

[Task: Open a laptop]

Spirit v1.6 result (left) vs. Cosmos3 result (right)

[Task: Place toy in plate]

Spirit v1.6 result (left) vs. PI0.5 result (right)

This top ranking is no one-off breakthrough, but the accumulated result of Spirit AI's long-term deep cultivation and high-frequency iteration in embodied models. Over the past six months, from the leapfrog evolution of Spirit v1.5 to the new-generation Spirit v1.6 once again claiming the top spot on an authoritative international benchmark, Spirit AI has used an industry-rare "consecutive championship" record to validate the forward-looking nature of its systematic, industrialized approach to technology development. This sustained, stable generational lead not only highlights the team's profound technical depth in general-purpose embodied foundation model R&D, but also demonstrates Spirit AI's formidable capability to continuously translate frontier breakthroughs into industrial-grade, systematic technical assets.

Data Wins: Building a "Data Pyramid" and a Million-Hour Core Moat

Behind the generational evolution of embodied foundation models, beyond extreme engineering at the algorithm level, lies Spirit AI's solid data infrastructure. The generalization capability of embodied intelligence is fundamentally determined by high-quality, large-scale real-world data.

By 2026, Spirit AI will have accumulated a milestone of 1 million hours of real-world interaction data, encompassing extremely rich long-tail complex scenarios and multi-modal operation samples.

It is precisely this underlying data infrastructure with exceptionally high "data density" and "scenario purity" that provides uninterrupted data fuel for the generalization and evolution of the Spirit v1.6 model, building a core moat of lasting competitiveness for Spirit AI in the embodied foundation model race.

Fundraising at Full Speed: Nearly 5 Billion Yuan in 3 Months Sets Industry Record

Alongside high-frequency technology and product iteration, Spirit AI has demonstrated unmatched magnetic pull in capital markets. In February 2026, Spirit AI completed two consecutive rounds totaling nearly 2 billion yuan, with valuation exceeding 10 billion yuan. In April, it closed another 1 billion yuan round, jointly led by Shunwei Capital and Yunfeng Capital. Just one month later, it completed another 1.5 billion yuan Series A+ round, with investors including top-tier dollar funds, major strategic investors, and state-backed funds, as existing shareholders continued to increase their stakes — assembling a premium capital lineup spanning top financial PE firms, international dollar funds, industrial strategic capital, and local government investment platforms.

Four密集 funding rounds in just three months, with cumulative funding nearing 5 billion yuan, set a new record for financing frequency in the embodied intelligence track.

Proceeds from this round will focus on three core directions: continued R&D iteration of next-generation embodied intelligence foundation models, construction of a global real-world data infrastructure system, and scaled commercial deployment across multiple industry scenarios. Leveraging full-stack technical advantages and capital support, the company will continue refining its closed-loop system of algorithms + hardware + data + scenarios + ecosystem, consolidating long-term competitive barriers.

Resonance with Giants: Strategic Alliances with Majors Unlock "Spirit AI Speed" in Commercialization

With the dual boost of capital and technology, Spirit AI has consistently adhered to a dual-wheel drive of R&D and industry application, rejecting conceptual navel-gazing in the sector. It is now advancing commercial deployment across massive scenarios at a rare "Spirit AI speed," precisely constructing a commercialization golden triangle spanning "global industry, China retail, and high-end manufacturing."

Through lightning-fast alliances and deep collaboration with industry benchmark majors including Bosch Group, JD.com, and CATL, Spirit AI has in extremely short order established a positive flywheel of "scenarios precipitate data, data iterates models, models feed back into industry." This not only validates with hard strength the genuine value of embodied intelligence empowering the real economy, but also lays an insurmountable first-mover advantage for subsequent cross-industry, systematic seamless scaled replication.

With a top-tier founding team, globally leading embodied foundation model technology, differentiated hardware barriers, scaled real-world data reserves, and a full-chain industrial cooperation ecosystem, Spirit AI continues to lead the domestic embodied intelligence track. Standing at a new starting point for industry development, the company will use this funding round and its global model ranking as a new springboard, continuing to focus on embodied foundation model R&D, building a million-hour real-world data system, and scaling deployment across all industry scenarios.