Meta's Muse Spark series, particularly the 1.1 and 1.3 variants, currently posts the strongest HLE results among its models, reaching 62.1% on select leaderboards and trailing Anthropic's leading Claude variants by roughly 3 percentage points. Rapid iteration on reasoning-heavy architectures and tool-use integration has driven recent gains, mirroring the broader 30-point HLE improvement across frontier systems over the past year. Competitive pressure from closed labs releasing frequent updates keeps Meta focused on closing capability gaps before year-end, though exact release timing and benchmark saturation risks introduce uncertainty. Traders watch for new Muse or Llama variants, developer conferences, and any official capability announcements that could shift Meta's position on the 2,500-question expert benchmark.