Anthropic Fable 5 Leads Humanity's Last Exam as OpenAI and Anthropic Restrict Top Models to Government-Approved Buyers
On the single cross-vendor benchmark where all three labs can be compared directly, Claude Fable 5 scores 53.3 percent on Humanity's Last Exam, ahead of GPT-5.6 Sol at 47.2 percent and Gemini 3.1 Pro Preview at 44.7 percent. Both OpenAI and Anthropic voluntarily restricted their highest-capability new models to government-approved customers during a 30-day pre-deployment window established by Executive Order 14409, meaning most enterprise buyers cannot access Sol or Mythos 5 at launch. The pricing spread is also consequential: GPT-5.6 Sol lists at $5 per million input tokens versus $10 for Anthropic's Fable 5 and Mythos 5, while the budget-tier GPT-5.6 Luna at $1 per million input tokens is the cheapest capable frontier model currently on public record.