Keep pulling the thread on Dario Amadei.
Leaders at Anthropic, including CEO Dario Amadei, reportedly believe the company might eventually be the only private company remaining, alongside governments.
ZAI's GLM-5.3 achieved state-of-the-art results on the AutomationBench and GDPVal benchmarks for agentic use cases, slightly outperforming US frontier models.
During the Hugging Face cyber attack, defenders used ZAI's GLM 5.2 because safety guardrails on frontier models made them ineffective for defensive tasks.
ZAI's GLM-5.3 improved its score on the CyberGym benchmark by 7 points compared to its predecessor, surpassing the performance of Fable-5.
The per-token cost of ZAI's GLM 5.3 is less than 10% of the cost of Fable or 5.6 SOL, and 20% of the cost of KIMI K3.
ZAI claims its models have detected over 1,000 critical and high-risk vulnerabilities in open-source software repositories.
Anthropic may not publicly release its next-generation model, opting to use it internally instead.
As of mid-July, Anthropic possessed an unreleased model, "Model 2", which it described as "somewhat more capable" than Mythos-5.
Anthropic's unreleased "Model 2" scored 62.8% on the internal CoBenchMark v2, compared to 50.3% for Mythos 5 and 54.8% for Mythos Preview.
Anthropic has started meeting with investors and investment banks in preparation for a potential IPO.
Anthropic's revenue in Q2 reached $11.5 billion, a 14-fold increase year-over-year.
Investors expect Anthropic to achieve a valuation of $2 trillion in its initial public offering.