Claude Mythos 5: Anthropic's 10-Trillion-Parameter Bet on AI Safety
By Sanna the Weaver • Sat Mar 07 2026 • Technology
Anthropic, the AI safety company founded by former OpenAI researchers, has unveiled Claude Mythos 5 — with 10 trillion parameters, it is the largest AI model ever publicly announced by any company. The announcement comes alongside Capabara, a mid-sized model targeting specialized applications in cybersecurity, coding, and ethical AI development. Together, they represent Anthropic's most ambitious push yet to demonstrate that safety and capability are not in conflict. The Scale Ten trillion parameters is a number that requires context. GPT-4, released in 2023, is estimated to have approximately 1.8 trillion parameters. Claude Mythos 5 is roughly five times larger by this measure. But raw parameter count is an imperfect proxy for capability — the architecture, training methodology, and the quality and curation of training data matter as much as scale. What makes Mythos 5 significant is not merely its size but Anthropic's claims about what that scale enables: genuine multi-step reasoning across complex domains, dramatically reduced "hallucination" rates on factual tasks, and improved performance on tasks requiring sustained logical coherence over long passages. Constitutional AI at Scale Anthropic's signature contribution to AI development is Constitutional AI — a training methodology in which the model learns not just from human feedback on its outputs but from a set of principles it applies to critique and revise its own responses. As Mythos 5 scales, Anthropic claims that constitutional alignment becomes more robust rather than less — a claim that challenges the assumption, common in AI safety research, that larger models are inherently harder to align with human values. Independent evaluations are ongoing. "Scale is not the enemy of safety. Thoughtless scale is. We have been thinking about this for three years." — Anthropic CEO Dario Amodei, February 2026 Revenue and Competition Anthropic's annualized revenue has reached approximately
9 billion, driven by Claude's adoption in enterprise software development, legal research, and healthcare. This places it in direct competition with OpenAI, which has surpassed $25 billion in annualized revenue and is taking early steps toward a public listing. The AI model market is consolidating rapidly around a small number of frontier players — OpenAI, Anthropic, Google DeepMind, Meta, and xAI — each with models that are beginning to differentiate less on capability and more on deployment ecosystem, pricing, and trust. Anthropic is betting that "trustworthy AI" is a durable competitive advantage, not merely a marketing claim.