Anthropic Launches Claude Opus 5.5, Cheaper Model Beats Fable 5.1 on Coding Benchmarks

Anthropic unveiled Claude Opus 5.5 on September 22, 2026, reducing prices by 40% while outperforming its own Fable 5.1 on agentic coding and knowledge tasks. The new flagship model costs $4 per million input tokens and $20 per million output tokens—20% cheaper than Opus 5—with inference speeds improved by over 30% percent. On Terminal-Bench 4.0, an industry standard for agentic command-line work, Opus 5.5 reaches 66.4% accuracy, outpacing Fable 5.1's 55.8%. On professional white-collar benchmarks like GDPval-AA, the model achieves higher Elo ratings than competitors. Input costs also fell 60% on cached tokens, dropping from $0.50 to $0.20 per million.