It's a total explosion! The super giant suddenly strikes! Another trillion-dollar opportunity emerges

For stock trading, just look at the Golden Qilin analyst research reports—authoritative, professional, timely, and comprehensive. Help you uncover high-potential theme opportunities!

Recently, news that OpenAI invested in AI startup Isara has completely ignited global attention in the AI community on large-scale multi-agent collaborative architectures.

Isara’s core selling point is only one: its large-scale multi-agent collaborative architecture. Its founder defines the technology as: a paradigm revolution from an “isolated tool” to a “collaborative team.”

The multi-agent track is the next trillion-level AI boom. OpenAI’s move above, in essence, is a clear bet by global AI giants on the next-generation technical direction in AI.

As the only multi-agent track player in the China market, Shanghai Tigrad Echo Intelligent Technology Co., Ltd. (hereinafter referred to as Tigrad Echo, English name: Gradence) is a key and scarce target for domestic capital to deploy at the global AI frontier.

WeCode delivers double first place

Multi-agent models outperform a single-agent flagship model

The moment the news that OpenAI invested in Isara broke, the global AI community was completely set on fire. Investors “vote with their feet,” concluding that marginal improvements in single-model capabilities have already peaked—multi-agent clusters are the next stop on the path toward AGI (artificial general intelligence).

Tigrad Echo’s core product WeCode (multi-agent on Codex) ranks first on the SWE-bench-verified leaderboard, with a problem-solution success rate of 86.90%.

By comparison, second place is agent+Opus 4.6 under Anthropic, with a problem-solution success rate of 80.80%; third place is OpenAI’s GPT-5.2, with a problem-solution success rate of 80.00%.

This means Tigrad Echo’s WeCode leads the second-place player by 6.1 percentage points, and it is also the only AI product solution globally to break through the 85% threshold.

Meanwhile, on the higher-difficulty SWE-bench Pro (a difficult complex task set), WeCode also claimed the global first place, surpassing Opus4.6, GPT-5.4, and other flagship models based on a single agent.

In other words, multi-agent track players such as Tigrad Echo have proven the generational advantages of multi-agent collaborative architectures—able to work together like a well-trained team of programmers, rather than a “solo fighting” coding assistant.

The SWE-bench leaderboard is released by Princeton University and is one of the most authoritative benchmarks for AI programming ability globally. The leaderboard fully simulates real software engineering scenarios: it gives AI a GitHub (software project hosting platform) open-source project’s software defects, and requires automatically generating code patches that can pass all tests.

“Same-house” showdown in the multi-agent track

The China route is more forward-looking than Isara

A coincidence is that Tigrad Echo’s technical roadmap is highly aligned with Isara—OpenAI’s just-bet target—both focus on multi-agent collaboration, but a careful comparison shows there are clear differences between the two.

Isara’s core positioning is large-scale intelligent agent orchestration, solving the problem of “multiple agents working together on the same tasks.” It is still in the architecture validation phase and has no publicly disclosed, deployed benchmark results.

Tigrad Echo goes one step further, focusing on “multi-agent deep collaboration + granting agents autonomous awareness.” Its programming agent, WeCode, has completed a technical closed loop in AI programming scenarios, secured double top rankings worldwide, and can be quickly expanded into industries such as finance, research, and manufacturing.

According to publicly available information, WeCode is built on a foundation of general-purpose large models to construct a multi-agent collaboration system. It achieves self-organizing and flexible collaboration through an underlying collaboration network model and a swarm intelligence learning algorithm. In ultra-complex task scenarios, it significantly improves task completion rates.

An “underestimated” AI field “golden target”

One comparison worth noting is this: Isara currently has no deployed product or publicly available benchmark results, and its valuation has reached $650 million; meanwhile, Tigrad Echo holds double worldwide first-place rankings on both SWE-bench Verified and SWE-bench Pro—its product is already running in practice, and its technical roadmap is also more forward-looking.

As a representative company in China that has global competitiveness in this track, Tigrad Echo is an important scarce target for domestic capital to deploy in the global AI frontier.

Multi-agent collaboration is widely considered to be one of the key paths to AGI and is expected to become the core growth direction in the next stage of the AI field. Tigrad Echo’s technical accumulation lays the foundation for China to secure a favorable position in the AGI era.

In the past few years, China’s AI industry has repeatedly been criticized for “lively applications at the application layer, while catching up on underlying technologies.” But the results of WeCode on the SWE-bench leaderboard are rewriting the logic of the established narrative.

Challenges still exist, but one fact can no longer be ignored—when Silicon Valley started going crazy with bets on the concept of “multi-agents,” China had already produced a leading contender with global competitiveness.

		Sina Statement: This message is reprinted from media partners of Sina. Sina.com publishes this article for the purpose of transmitting more information, and does not mean agreeing with its viewpoints or confirming the descriptions therein. The article content is for reference only and does not constitute investment advice. Investors act at their own risk.

Massive information, precise interpretation—everything is in the Sina Finance APP

Responsible editor: Guo Xutong