Tencent Hyqian released the research Agent Hyra 2.0 and disclosed multiple R&D achievements

AT A GLANCE

Tencent disclosed that Hyra 2.0 can carry out scientific research tasks over the long term and has already participated in the R&D of models, chips, and robots.

Article

Tencent Hyra 2.0 Advances AI-Driven Independent Research: Training Models from Scratch, Designing Robots, and Even Participating in the Development of Next-Generation AI

Beating AI Flash News: Tencent Huan Yuan has released its research Agent Hyra 2.0, shifting the research focus from repeatedly trial-and-erroring at individual problems to developing scientific models, algorithms, and tools that can be reused over the long term.

To enable AI to conduct research continuously for days or even weeks, Hyra 2.0 divides tasks among three layers of Agents—Overall, Mission, and Worker—responsible for resource coordination, research planning, and experiment execution, respectively. Tasks are persistently saved, allowing other research to proceed while awaiting training results. Each conclusion must be accompanied by evidence such as code and logs, and submitted to independent Agents for verification when necessary. Valuable experience is also distilled into Skills for use in subsequent tasks.

Tencent disclosed that Hyra 2.0 has trained two protein models from scratch. Among them, HyInv with 1,960,000 parameters can design protein sequences based on target scaffolds, generating at a speed 57 times faster than the commonly used ProteinMPNN. In warm-up tests on a single H20 GPU, generating 128 sequences of approximately 400 residues takes only about 50 milliseconds. The other model, HyPLM-Flash, has 650,000,000 parameters and was trained on 726,000,000,000 tokens. Weights, inference code, and technical reports for both models have been made publicly available.

Hyra 2.0 also participated in optimizing the open-source RISC-V processor "Xiangshan," improving SPEC CPU2006 performance per GHz by 5.79% while increasing chip area by only 0.3%.

It also designed a "Spiral Ladder Hand" with five fingers of gradually increasing length. Initially about 11% slower than the original optimal design, after training with a control algorithm under the same budget, its simulated rotation speed improved by 28.6%.

Currently, Hyra 2.0 is also involved in the pre-training, post-training, and inference system optimization of the next-generation Huan Yuan models, aiming to enable AI to help train stronger next-generation AIs by improving development tools.

Original article link https://m.theblockbeats.info/flash/375029

Key points

Tencent Hyqian has released the research Agent Hyra 2.0, expanding its research focus to long-term reusable scientific models, algorithms, and tools.
Hyra 2.0 assigns tasks to the three-layer Agent system of Overall, Mission, and Worker, and requires conclusions to be supported by evidence such as code and logs, with independent Agents conducting reviews when necessary.
Hyra 2.0 has trained two protein models, HyInv and HyPLM-Flash, from scratch.
HyInv has 1960000 parameters and generates 57 times faster than the commonly used ProteinMPNN; in a single H20 GPU warm-up test, generating 128 sequences of about 400 residues takes about 50 milliseconds.
HyPLM-Flash has 650000000 parameters and was trained on 726000000000 tokens; the weights, inference code, and technical report for both models have been made public.
Hyra 2.0 participated in optimizing the open-source RISC-V processor "Xiangshan," improving SPEC CPU2006 performance per GHz by 5.79% while increasing chip area by 0.3%.
Hyra 2.0 designed a "spiral staircase hand" with five fingers of increasing length; after training the control algorithm with the same budget, the simulated rotation speed was 28.6% faster than the original optimal design.
Hyra 2.0 has participated in the pre-training, post-training, and inference system optimization of the next-generation Hunyuan model.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

Hyra 2.0 is a research Agent system that breaks down long-term research into tasks such as planning, coordination, and experimental execution, while preserving task and experimental evidence. The materials state that it has been used for protein modeling, processor optimization, robot design, and the development of next-generation hybrid models.

Why it matters to readers

If the results in the material can be reproduced, Hyra 2.0 demonstrates the direction of AI expanding from executing a single experiment to continuously managing the research workflow.

For newcomers, the key is not to memorize all model names, but to understand: the system delegates tasks, preserves the research process, requires evidence, and distills effective experience into reusable Skills for future use.

Risks and unknowns

The provided material does not include a link to Tencent’s original official announcement; the information is primarily derived from media briefings.
The independent reproducibility of the model performance, processor performance, and robot simulation results described in the paper is unknown.
Hyra 2.0's specific contributions to and actual benefits from participating in the research and development of the next-generation Hunyuan model have not been disclosed.

Tencent Hyqian released the research Agent Hyra 2.0 and disclosed multiple R&D achievements

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center