Anthropic’s April 2026 release of Claude Mythos Preview established it as a frontier large language model with exceptional agentic coding and cybersecurity benchmarks, yet safety restrictions have kept the full model out of public arenas. Recent June announcements of the safer Claude Fable 5 (now topping several coding leaderboards near 1525–1665) and limited Mythos 5 access for Glasswing partners signal strong underlying capability, but traders note that true Mythos deployment on Arena.AI’s coding leaderboard remains gated pending efficiency and trusted-access expansions. Competitive pressure from Opus variants and the model’s documented zero-day discovery strengths create expectations of a high debut score if released, though delays or further safeguards could push timelines into late 2026. Key near-term catalysts include any broader trusted-program rollout or official leaderboard appearance announcements.