Pré-encomendaLoja

Notícias Quânticas: segunda-feira, 15 de junho de 2026

Three stories this week, and they all point at the same thing: the field has stopped bragging about physical qubit counts and started talking about logical ones — the error-corrected qubits you can actually compute with. QuEra put the clearest stake in the ground, announcing 'Libra,' a fault-tolerant neutral-atom machine targeting more than 256 logical qubits at a 10⁻⁶ error rate, set to land on AWS Braket in 2028. The number to watch isn't 256, it's the 2028 — that's a roadmap, not a result, and while QuEra's existing hardware (Aquila, Gemini) is real, the fault-tolerant part is still on paper. IonQ, by contrast, showed a result rather than a roadmap: running nine error-correcting codes on a 40-ion barium chain, they pushed a logical qubit past 'breakeven' — the logical memory lasted about 3.95 seconds, longer than the physical qubits underneath it, with logical error rates four to nine times lower than comparable superconducting demos. Breakeven is the unglamorous line that separates error correction that helps from error correction that hurts, and crossing it on trapped ions is a genuine milestone. And IBM did something sideways but clever: they pointed an LLM-guided search (an open-source tool called OpenEvolve) at the problem of finding good quantum codes and turned up 465 candidate qLDPC codes — including a [[288,50,8]] code that packs 50 logical qubits where the prior best for that family was 16. The codes still have to be verified and built, but searching the design space with AI instead of hand-deriving codes one paper at a time is a real shift in how the work gets done. The through-line: nobody serious is selling raw qubit counts anymore. The whole conversation has moved to whether the qubits can be trusted — which, for anyone trying to learn what a quantum computer will actually do, is exactly the right question.