GPT-5.4 rates Millennium Prize difficulty, eyes 2027

Community speculation meets real milestones: GPT-5.4 recently solved a 20-year research problem on its 11th attempt, but Millennium Prize problems may require inventing new branches of math, not just scaling compute.

• –GPT-5.4 achieved 38% on FrontierMath Tier 4 problems and solved a decades-old problem by finding a preprint the human author had never encountered — literature archaeology is a genuine AI edge
• –FrontierMath benchmarks show ~25x improvement in 16 months, from under 2% to 50% on mid-tier and 38% on hardest problems
• –Experts including Terence Tao caution that Millennium problems require novel mathematical frameworks that don't yet exist — scaling context windows won't substitute for conceptual invention
• –The post's "10x time horizon solves Millennium problems" claim is speculative extrapolation, not a research finding — treat accordingly
• –Low-engagement source (score: 0, 19 comments) — this is community speculation, not an OpenAI announcement