Will an Anthropic Claude model score at least 45% on Humanity’s Last Exam?

Prediction markets put the probability at 20%: Will an Anthropic Claude model score at least 45% on Humanity’s Last Exam. Currently, markets see this as unlikely (20% YES). + Insurance Times publishes full Five Star Rating Report: MGA market for 2025/26.

What’s Happening

Anthropic announced on May 6, 2026 a compute partnership with SpaceX that grants Claude access to the Colossus data center, alongside additional capacity deals that allowed the company to raise API rate limits for Claude Opus models and expand Claude Code usage tiers. The agreement marks a notable shift given SpaceX chief Elon Musk previously labeled Anthropic "Misanthropic" and "evil," with xAI now folded into SpaceX. The expanded compute envelope is directly relevant to whether an anthropic claude model score at least 45% on humanity's last exam, since frontier benchmark performance has historically tracked training and inference scale. [Anthropic, May 6]

On May 5, 2026, Insurance Times reported that Anthropic would not release its most powerful model, Claude Mythos, to the general public, citing global cyber security concerns. The decision has rippled through cyber insurance underwriting, where unreleased frontier capabilities complicate risk modeling. The non-release does not by itself preclude internal evaluation runs on Humanity's Last Exam, the closed-ended academic benchmark covering graduate-level questions across mathematics, sciences and humanities, but it constrains the pool of publicly verifiable Claude variants from which a qualifying 45% score could be confirmed before resolution. [Insurance Times, May 5]

Commercial momentum continues in parallel: on May 4, 2026, Anthropic announced a $1.5 billion joint venture with Goldman Sachs, Blackstone, Hellman & Friedman, Apollo Global Management and General Atlantic to deploy Claude across PE-owned mid-sized companies. Three days earlier, on May 1, 2026, CNN reported the Pentagon awarded contracts to seven Big Tech firms while excluding Anthropic over disagreements on AI warfare safety guardrails. The split commercial-versus-defense posture frames the operating environment in which an anthropic claude model score at least 45% on humanity's last exam would need to be demonstrated. [CNBC, May 4]

Traded on Polymarket — $163K Volume

Polymarket prices this at 48c YES with $163K in volume. Moderate liquidity — use limit orders for positions above $1K to avoid moving the price.