Whoops! Sorry about that.

TL;DR: Newer LLMs are not necessarily better at domain-specific tasks. On our fraud benchmark, newer Opus, Sonnet, and GPT versions all had lower recall, F1, and dollar-weighted recall—even when precision improved. The lesson: choose models using high-quality, domain-specific benchmarks, with reproducible evaluation and ideally deterministic scoring against verified outcomes. These findings also motivated our next step: post-training open-weight models for greater control over quality, cost, and deployment.

L'integrazione di Deribit è completa e, per la prima volta nella storia del mercato, Coinbase Financial Markets collegherà a breve i derivati sulle criptovalute statunitensi e globali in un unico pool di liquidità regolamentato.

TL;DR: How Coinbase built, evaluated, and post-trained an Onramp fraud agent to achieve 9.6% higher performance (F1 score) and 55% lower end to end latency than frontier models.

TL;DR: We moved more than 300 Temporal namespaces, across more than 300 services, from a cluster we ran ourselves onto Temporal Cloud. You cannot finish the move by repointing services at Temporal Cloud while workflows are still running on the old one. Where teams could pause new work, we waited for in-flight runs to finish and then switched; for trading, payments, and long-running work, we ran the old cluster and Temporal Cloud, and moved new workflows over in steps, with rollback as a config change. Private connectivity, per-namespace service allowlists, and encrypted payloads were in place before the first production namespace moved.

TLDR: In October, Brazilians head to the polls. Tens of millions of votes will be cast by crypto holders. Cryptocurrencies are already part of people’s lives, with nearly 29 million Brazilians having or having previously owned digital assets and 65.8% of investors consider it important to vote for lawmakers committed to protecting the rights of digital asset users.