Updated for Week 39, 2026

Claude Fable 5.1 vs GPT-6 Astra

On our Week 39, 2026 scores, Claude Fable 5.1 edges ahead (95 vs 94). Pick Claude Fable 5.1 for the hardest reasoning problems. Pick GPT-6 Astra for computer use and automation.

Claude Fable 5.1GPT-6 Astra
MakerAnthropicOpenAI
ReleasedSeptember 1, 2026—
Price / 1M tokens$10 in / $50 out$10 in / $50 out
Open weightsNoNo
Power Rankings rank#2 (95)#3 (94)
Coding rank#3 (93)#2 (94)
Best for
  • The hardest reasoning problems
  • Long-running agents
  • High-stakes research
  • Computer use and automation
  • Scientific and terminal workflows
  • Professional documents
Strengths
  • #1 on the Arena agent leaderboard
  • Cache reads cut to $0.25 per million tokens
  • Leads AutomationBench and Terminal-Bench-Science
  • Scored 53 on the Artificial Analysis Intelligence Index
Watch out for
  • Premium price: 2.5x Opus 5.5
  • Shipped with breaking API changes
  • Same premium price as Fable 5.1
  • Sol now brings most of its gains at a fifth of the cost

The short version

Claude Fable 5.1: Anthropic's largest publicly available model. Fable 5.1 is priced at the premium tier and leads the Arena agent leaderboard.

GPT-6 Astra: OpenAI's top-tier GPT-6 model. Astra's training methods were reused for the cheaper Sol and Luna tiers, and it still leads a handful of automation benchmarks.

Rankings change every week. Check the latest standings or swap in a different model.