Anthropic Ships Claude Opus 5.5 at 20% Lower Token Prices
TL;DR
- The headline 40% cost reduction is skewed by a 60% cache-read cut, which dominates actual spend in agentic and coding workloads.
- Price-adjusted, Opus 5.5 scores 54.6% on FrontierCode at roughly one-fifth the cost of GPT-6 Astra's top score on the same benchmark.
- Anthropic has flagged Sonnet 5.5 and Haiku 5.5 arriving in coming weeks, meaning the full lineup repricing lands before the reported November IPO.
Anthropic's Claude Opus 5.5, released September 22, lists at $4 per million input tokens and $20 per million output. That is 20% below Opus 5. Cache reads drop 60% to $0.20 per million, which the company notes make up the majority of agentic and coding work costs. Anthropic puts the overall savings at roughly 40% on typical workloads.
On its own benchmarks, Opus 5.5 posts 66.4% on Terminal-Bench 4.0 (up from Opus 5's 52.3%) and 81.8% on OSWorld 2.0 for computer use. The launch post says: "On Terminal-Bench 4.0, it matches Astra for about 40% of the cost, while on CursorBench it beats GPT-5.6 Sol by 11 points for about a third of the cost." Output runs "more than 30% faster than Opus 5."
Anthropic offers one customer anecdote: "One tester completed a 680,000-line code migration in less than a day."
The safety numbers are the other pitch. On a benchmark run by AI security firm Gray Swan, Opus 5.5 ties Claude Fable 5.1 for the lowest prompt-injection success rate of any model tested. On Anthropic's internal containment eval, the model "attempted to circumvent boundaries around 85% less often than Opus 5 or Claude Mythos 5.1." Cybersecurity queries route to Opus 4.8, and biology research requires approval through Anthropic's Life Sciences Verification Program. Pre-release evaluations came from Frontier Design and METR.
The model ships across AWS, Google Cloud, and Azure, with Sonnet 5.5 and Haiku 5.5 due in the weeks ahead. It is the fourth Anthropic item on our tracker today.
What others are reporting
-
TechCrunch Read →
First release since Amodei's Sept 12 'pace the frontier' essay; frames the launch against Anthropic's safety pledge and names METR and Frontier Design as external evaluators.
the strongest-performing model we've tested to date
-
The Decoder Read →
Focuses on Anthropic's commitment to reduce Claude's formulaic style, framing the writing quality change as a user experience shift separate from benchmark gains.
Puts the most important information first, uses less jargon, and follows writing instructions more closely.
-
MarkTechPost Read →
Notes thinking is non-disableable, EU watermarking is mandatory, and cache reads dominate agentic costs; backs claims with the 680K-line migration case study.
Cache reads make up most agentic and coding costs, and they drop 60%.
-
FinOut Read →
Enterprise FinOps angle: argues cache hit rate swings real-world costs by 10x on identical workloads, with spend allocation and anomaly alerting guidance for procurement teams.
Cache hit rate is the single biggest lever on real-world Opus 5.5 spend, more than the headline per-token price.
-
MacRumors Read →
Positions the launch within Apple's ecosystem, citing Claude's existing CarPlay integration as context for why speed and cost gains matter for device-level deployment.
Opus 5.5 performs at the level of Claude Fable 5.1 on most work, but costs 40 percent less to run than Opus 5.
Shared on Bluesky by 5 AI experts
-
-
Tim Kellogg @timkellogg.me
Opus 5.5 better and cheaper than Opus 5 — typically 40% cheaper, $4/mtok input, $20/mtok out www.anthropic.com/claude-opus-...
View on Bluesky →
Originally reported by anthropic.com
Read the original article →Original headline: Anthropic Ships Claude Opus 5.5 at 40% Lower Cost, First Model Since Amodei's 'Pace the Frontier' Essay