Behold! The Machine That Works While You Sin: GPT-5.5 Arrives
The essence of GPT-5.5, they say, is its ability to “look at an unclear problem and figure out what needs to happen next.” Greg Brockman, the co-founder and president of this digital pantheon, proclaims it “way more intuitive to use,” a machine that carries the weight of labor with a grace that eludes even the most diligent of mortals. On Terminal-Bench 2.0, it scores 82.7%, surpassing Claude Opus 4.7, which lags behind at 69.4%. On GDPval, it achieves 84.9%, and on OSWorld-Verified, 78.7%. Amelia Glaese, the Chief Research Officer, declares it “definitely our strongest model yet on coding,” a testament to its prowess, or perhaps, a reminder of our own inadequacies. And yet, it operates with the same per-token latency as its predecessor, using fewer tokens to complete the same tasks. Efficiency, thy name is machine!


![The study demonstrates how constraints progressively refine predictions for the mass-radius relation of neutron stars, narrowing the range of possible equations of state-from those based solely on chiral effective field theory, to those incorporating maximum mass limits [latex]M_{max} \leq 2.16^{+0.17}_{-0.15}[/latex], and finally, those informed by astrophysical observations of pulsars like PSR J0030+0451 and gravitational wave events like GW170817, ultimately revealing which equations of state persist under increasing scrutiny.](https://arxiv.org/html/2604.21039v1/plots/CQR-last.png)

