Claude Opus 5.5 Launches at Fable-Level Performance for 40 Per Cent Less, With Sonnet and Haiku to Follow

Anthropic released Claude Opus 5.5 on Tuesday, the first model in what the company is calling its 5.5 family, and pitched it on a single trade: most of the capability of its own top-of-range model, Claude Fable 5.1, at a fraction of the price and more than a fifth faster to answer. Input tokens cost 4 dollars per million and output 20 dollars per million, a 20 per cent cut from Opus 5's rates, and Anthropic says the new model generates its answers over 30 per cent faster than its predecessor while running 40 per cent cheaper to serve. It is live today across the API, the Claude apps, Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry. MacRumors and 9to5Mac both carry the release-day figures against Anthropic's own announcement.

What actually improved

Anthropic's benchmark tables put Opus 5.5 ahead of Fable 5.1, its own pricier and slower model, on agentic coding, general knowledge work, computer-use tasks, visual chart and document recognition, and multidisciplinary reasoning; on other categories the two are close enough that the company describes Opus 5.5 as matching Fable 5.1 "on most work". That is an unusual message from a lab: normally the newest, most expensive model is presented as the best across the board. Here Anthropic is explicit that Fable 5.1 remains its frontier model for the hardest problems, and that Opus 5.5 is the one most customers, especially anyone running agentic coding workloads at volume, as we covered when Opus 5 first shipped, should actually be running day to day.

The pattern behind it

Anthropic has spent 2026 narrowing the price and latency gap between its middle and top tiers rather than only chasing frontier scores, a strategy this outlet has tracked through the Sonnet 5 pricing cliff and the debate over a two-tier access model that followed the Mythos-class launch. Opus 5.5 continues that: a mid-cycle refresh, not a renamed frontier model, aimed at enterprise customers who care more about a predictable bill and a fast reply than about winning the next leaderboard. Anthropic has said Sonnet 5.5 and Haiku 5.5 will follow within weeks, which would complete the same refresh across the whole range.

Context: a crowded fortnight for flagship models

The launch lands three weeks after OpenAI's own frontier release, GPT-6 Astra, which we report separately reached OpenAI's highest-ever internal reasoning and cybersecurity scores but which independent evaluators still rate behind Fable 5.1 on broad coding and intelligence indices. Neither company is claiming an outright win over the other on every measure, and the more interesting story for anyone actually building on these models is what Opus 5.5 signals about cost: a model priced a fifth below its predecessor and pitched as good enough for most production work is a bet that most customers do not need the absolute frontier, they need something reliable that does not eat the margin. Whether that bet is right will show up in Anthropic's revenue mix, not in a benchmark table, over the next two quarters.