Anthropic has shipped Claude Opus 5.5, and the headline claim is a compression story: the new model performs at the level of Claude Fable 5.1 — the restricted frontier tier the company usually holds back for cybersecurity and advanced research — on most everyday work, while costing 40% less to run than its predecessor. It was released on September 22, 2026, two months to the day after Opus 5.
The Performance Claim
According to Anthropic’s announcement, Opus 5.5 is “the first model in our new Claude 5.5 family,” and it is the new leading model in the company’s lineup on most work. The benchmark detail is where the pricing story emerges.
On FrontierCode v1.1 — a benchmark measuring whether an agent’s code changes would survive review — Opus 5.5 scored 54.6% at its default effort level, ahead of OpenAI’s GPT-6 Astra top score of 53.3%, at roughly 20% of the cost per task. On Terminal-Bench 4.0, it beats Opus 5 at max effort for about a fifth of the cost and matches GPT-6 Astra at about 40% of the cost.
The early-tester anecdotes follow the same pattern. One completed a 680,000-line code migration in under a day. Spotify reported matching Opus 5’s quality on agentic coding tasks in about half the turns and output tokens, cutting workload cost by 40 to 50%. Optiver said the model passed trading-support tasks earlier Claude models had failed.
The Pricing Story
Output tokens cost $20 per million, down from $25 on Opus 5. Input is $5 per million. The bigger move is on cache reads, which Anthropic says make up the majority of agentic and coding work costs: $0.20 per million tokens, 60% less than Opus 5. The model also generates output more than 30% faster, and a Fast mode in Claude Code and the Claude Platform runs at up to 2.5x speed for $8/$40 per million.
The compression matters against Anthropic’s own recent pricing history: the lab raised its premium subscription prices sharply this year while rivals cut theirs. Opus 5.5 is the counter-move — same work, cheaper — and it lands one day after OpenAI released GPT-6 Astra more widely. TechCrunch noted the launch came with Anthropic’s CEO having recently embraced calls to pace frontier development, making the cadence of this release notable: the model improves on its predecessor while the frontier itself is deliberately slowing.
The Alignment Claim
The more consequential number is not on a coding benchmark. On Anthropic’s automated behavioral audit — the alignment suite that tests Claude across thousands of simulated scenarios — Opus 5.5 achieved the best scores of any model the company has tested to date. It is less likely than recent models to take hard-to-reverse actions, act outside its assigned boundaries, or fall for prompt injection.
This continues the argument Anthropic has been making since Opus 5 launched with the lowest misalignment score the lab had recorded: that alignment improves with capability rather than trading off against it. Two data points in a row now support that claim — at least as measured by the company’s own audit, which has not been independently replicated.
The safety posture also changed in one important way. Anthropic says Opus 5.5 is comparable to Claude Mythos 5.1 in biology and cybersecurity capability, so it is being deployed with safeguards similar to those on Fable 5.1 — the tier Anthropic previously restricted because its capabilities justified it. A workhorse model crossing into restricted-tier capability is the quiet story here: the line between “everyday model” and “restricted model” is moving, and the safeguards are moving with it. Vetted organizations can apply to the Life Sciences Verification Program now, with Cyber Verification Program access expanding in coming weeks.
What It Means
Two months from Opus 5 to Opus 5.5, at lower cost, with better alignment scores and restricted-tier capability — that is not the profile of a lab slowing down. It is the profile of a lab that has decided its constraints are compute and safety evaluation capacity, not ambition. Sonnet 5.5 and Haiku 5.5 are promised “in the coming weeks,” which would complete the family refresh within weeks of the flagship.
The open question is whether the alignment claim survives outside scrutiny. Anthropic’s own audit produced the number, and the system card is available for review, but the company’s history of measuring itself — and competitors’ history of disputing those measurements — means the claim is best read as a strong opening position, not a settled result.
