Anthropic released Claude Opus 5.5 on Tuesday, September 22, 2026, billing it as the first entry in its Claude 5.5 family and pricing output tokens at $20 per million, down from $25 for Opus 5, according to Techcrunch. Anthropic says the model performs at the level of its priciest flagship, Fable 5.1, on most tasks.
The release lands about two months after Opus 5 arrived on July 24. Decrypt reported the model is 20% less expensive to run than Opus 5 and 60% cheaper than Fable 5.1, with input tokens at $4 per million versus $5 before.
Also read: Meta's Muse Beats ChatGPT's Early Downloads, Appfigures Says
Key facts
- Anthropic calls Opus 5.5 its strongest-performing model tested to date, saying it outpaces the larger Fable model in many benchmarks (Techcrunch).
- Output tokens cost $20 per million, input tokens $4 per million, and cache reads fall 60% to $0.20 per million (Decrypt).
- On Terminal-Bench 4.0, Opus 5.5 scored 66.4% against Fable 5.1’s 55.8% and GPT-6 Astra’s 57.9% (Decrypt).
- Anthropic says Sonnet 5.5 and Haiku 5.5 will follow in the coming weeks carrying the same price cuts down the lineup.
- Opus 5.5 launches with the same cybersecurity and biology safeguards as Fable 5.1, with most cybersecurity tasks rerouted to the older Opus 4.8 (Decrypt).
Where Opus 5.5 leads and where it does not
On Anthropic’s own published numbers, Opus 5.5 leads Fable 5.1 and Opus 5 on coding and knowledge-work tests. Decrypt reported GDPval-AA v2.1, which grades professional work across 44 occupations using Elo, put Opus 5.5 at 1846 against Fable 5.1’s 1735 and Opus 5’s 1708. FrontierCode, which checks whether an agent’s code changes would be merged in a real engineering pipeline, scored Opus 5.5 at 54.4% versus 50.3% for Fable 5.1.
The lead is not complete. GPT-6 Astra’s 41.4% on AutomationBench, a Zapier-built test of full business workflows, edges out Opus 5.5’s 40.0%, and Astra’s 64.6% on Terminal-Bench-Science 0.1 beats Opus 5.5’s 58.7% by a wider margin. Anthropic concedes the gap between Fable and Opus is narrower than its benchmark scores suggest, though it notes Opus 5.5 is publicly available on plan and cheaper per token.
Also read: Petlibro Granary 2 Feeders Add AI Camera and Built-In Scale
Pricing, pacing and the Fable 5.1 backdrop
Techcrunch reported the model runs faster, reflecting lower compute requirements, and changes how Opus communicates — less jargon, with important information placed at the start of messages. The Decrypt account adds that Opus 5.5 is the most advanced model by version number (5.5 against Fable’s 5.1) and by family, with Opus the largest publicly available model, followed by Sonnet and then Haiku.
Anthropic acknowledged the reports’ shared context: Opus 5 shipped in July underpricing and outscoring Fable 5 on most benchmarks, Fable 5.1 arrived in early September and reversed that, and Opus 5.5 closes the gap again on price rather than raw scores. Lovable, a developer platform that ran Opus 5 through its own coding tests in July, said in a statement shared by Anthropic that the earlier model was steadier with far less variance run to run. Techcrunch noted the launch is Anthropic’s first since CEO Dario Amodei embraced calls to pace the frontier, quoting him saying he had become convinced that fully addressing the risks requires even more prudence, and that capability advancement should be paced so risk prevention can keep up.
Decrypt reported Amodei’s essay asked the industry to slow the pace at which it improves AI model capabilities. Safety training for Opus 5.5 was broadly similar to its predecessors, with alignment testing and pre-release evaluation by outside organizations including METR and Frontier Design, though Anthropic said more advanced training and evaluation systems, improved security and monitoring among them, are already being prepared. Its blog post argued public policy should play a larger role in keeping systems safe and said the company expects to share more details soon.
Why it matters
Opus 5.5 is live now across Anthropic’s platforms, including Amazon Web Services, Google Cloud and Microsoft Azure. Because it matches Mythos-class capability on biology and cybersecurity, safeguards limit how far the model can be used to find exploits in compiled programs or develop recognizable biological weapons. The combination means enterprise and developer buyers get Fable-tier performance at a materially lower per-token cost, which is the practical decision point for teams running long agentic coding sessions where cache reads dominate spend.
What to watch
Sonnet 5.5 and Haiku 5.5 are the next concrete milestones, with Anthropic saying they will arrive in the coming weeks and carry the same price cuts down the rest of the lineup. Anthropic also said further detail on its safety infrastructure and policy efforts is coming.
Sources: TechCrunch, Decrypt
