Claude Opus 5.5 is the model we would hand a messy, high-stakes job and trust to come back with the right answer. Anthropic says it performs at the level of its premium Fable 5.1 on most work, yet it costs less than half as much per token — and it now writes like a clear-headed colleague instead of a nervous graduate student.
Highest score on the independent Artificial Analysis Intelligence Index at launch (58, max effort). Leads GDPval-AA v2.1 real-world knowledge work at 1846 Elo, ahead of Fable 5.1 (1735). Token prices fell 20% to $4/$20 and cache reads fell 60% to $0.20, which Anthropic says makes typical workloads about 40% cheaper than Opus 5. Output is more than 30% faster, and five-hour usage limits went up on paid plans.
At maximum effort it thinks out loud a lot — Artificial Analysis measured roughly 119,000 output tokens per task against about 27,000 for GPT-6 Astra — so the low sticker price only pays off if you stay on the default effort for routine work. There is no image generation, reasoning can no longer be switched off, and most cybersecurity requests are quietly handed to the older Opus 4.8.
