Tencent Hunyuan launches Hyra 2.0 automated research framework

The framing matters more than any single benchmark: an agentic system that builds and upgrades reusable scientific models shifts AI's role from problem-solver to toolmaker, which could compound progress across protein design and quantum error correction rather than delivering one-off results. The claimed margins over established baselines like ProteinMPNN and the qLDPC Challenge leader, plus mathematician-verified proofs, raise the bar for independent replication. Whether outside groups reproduce these gains, and whether the reusable-tool approach generalizes beyond these domains, is the open question.