Anthropic releases Claude Opus 5.5, the first Claude 5.5-family model

Anthropic released Claude Opus 5.5 on September 22, 2026, the first model in a new Claude 5.5 family that the company says matches its own Claude Fable 5.1 on most work while costing 40% less to run than Opus 5.

What's new

Anthropic frames Opus 5.5 as a direct successor to Opus 5 rather than an incremental patch. In its own words: "Opus 5.5 is a major step up from Opus 5. It's the new leading model." The company points to a jump on Terminal-Bench 4.0, a benchmark for agentic terminal use, from 52.3% for Opus 5 to 66.4% for Opus 5.5, alongside a claimed 30% faster output generation.

Pricing dropped across the board versus Opus 5:

• Input tokens: $4 per million (down from $5)
• Output tokens: $20 per million (down from $25)
• Cache reads: $0.20 per million (down from $0.50, a 60% cut)

Anthropic says the combined effect is a "40% drop in costs" for typical workloads compared to Opus 5 — a notable move given Opus 5 itself launched as a cheaper alternative to Anthropic's own top-tier model.

On safety, Anthropic says "Opus 5.5 achieves the best scores of any model to date on our automated behavioral audit" and describes it as more resistant to prompt injection than Opus 5, calling it the first Opus model to launch with safeguards comparable to Fable 5.1 — previously Anthropic's most heavily hardened model. The company also says the model "communicates more naturally than prior models," with clearer writing over long sessions.

Context

Opus 5 itself launched by closing the gap on Anthropic's own flagship at roughly half the price, part of a broader industry pattern this year of frontier labs using price cuts and mid-cycle refreshes to hold ground rather than waiting for a full version bump. Opus 5.5 continues that pattern: rather than a new top-of-line model, Anthropic is repositioning its existing Opus tier to perform closer to Fable 5.1 while undercutting Opus 5 on cost — effectively pushing capability down through the price tiers.

The release also lands the same week Anthropic disclosed a partnership with Accenture to embed independent evaluators in its AI development process and extended its Compliance API to Claude in Chrome session transcripts, part of a broader September push on both capability and governance.

Why it matters

A 40% cost cut paired with a double-digit benchmark jump on agentic tasks is a meaningful shift in the price-performance curve buyers evaluate mid-tier models against, particularly for teams running Opus at volume for coding or terminal-agent workloads where Terminal-Bench performance is a direct proxy for real-world reliability. The safety claims are notable in their own right: Anthropic explicitly positioning Opus 5.5 as its most prompt-injection-resistant Opus model yet, with Fable-5.1-level safeguards, signals the company is treating agentic robustness as a competitive axis alongside raw capability and cost — not a separate track.