Our most expressive model yet: one million tokens of context, 38 ms to first token, and a voice you can tune in an afternoon.

A field guide to how we cut latency by 71% without trading away a single point on our quality evals.

How brand teams are replacing thousand-word prompts with a single living document the model actually reads.

Benchmarks measure correctness. We built a panel of poets, editors and support agents to measure something harder.

Forty people, three models, one very long table. What we argued about, what we agreed on, and what we shipped the week after.