Anthropic released Claude Opus 5.5 on 22 September 2026 with a price a fifth below its predecessor and a claim that reaches above it. The model "performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5," the company wrote. Fable 5.1 lists at USD 10 per million input tokens. Opus 5.5 costs USD 4.
For teams that moved demanding agent work to Fable, that is the offer: similar results for well under half the input price, on Anthropic's own numbers. Fable's restrictions come with it. Opus 5.5, the first model of a new Claude 5.5 family, inherits safeguards that send most cybersecurity work to an older model and screen biology requests, with verification programmes for professionals who need more.
- What: Claude Opus 5.5, first model of the Claude 5.5 family, API name claude-opus-5-5.
- Price: USD 4 and USD 20 per million input and output tokens; cache reads USD 0.20.
- Why it matters: Anthropic claims Fable 5.1 level results on most work at Opus prices, under Fable style safeguards.
Where the 40 percent comes from
"Opus 5.5 requires less compute to serve than Opus 5, and its pricing reflects that," Anthropic wrote in its announcement. Input and output tokens cost 20 percent less than Opus 5's USD 5 and USD 25. The deeper cut is in cache reads, the price of reading back a stored prompt, which the company says make up the majority of costs in agentic and coding work. They fall to USD 0.20 per million tokens, 60 percent below Opus 5. The 40 percent saving is Anthropic's test result for typical workloads at default settings, and the company says the model also writes output more than 30 percent faster.
Batch jobs cost half, USD 2 and USD 10, the documentation says. A fast mode with up to 2.5 times the speed costs USD 8 and USD 40 per million tokens. The announcement places it in Claude Code and the Claude Platform; the documentation calls it a research preview on the Claude API only, not on Amazon Bedrock, Google Cloud or Microsoft Foundry.
The benchmark figures are vendor reported. In Anthropic's table, Opus 5.5 scores 66.4 percent on Terminal-Bench 4.0, a test of agentic coding in a terminal, measured at its xhigh effort setting, against 55.8 percent for Fable 5.1 and 52.3 percent for Opus 5.
Fable's safeguards move down a tier
The rules came with the capability. "Because Opus 5.5 has extremely strong cyber capabilities, we're applying cybersecurity safeguards to Opus 5.5 that are similar to Fable 5.1's," Anthropic wrote. "Users will be able to identify and fix bugs in their code as part of the routine software development lifecycle, but most cybersecurity tasks will be re-routed to Opus 4.8." Opus 5.5 also uses the same biology safeguards as Fable 5.1.
In the API this shows up as refusals. The model runs a biology classifier next to the cybersecurity one, and a declined request comes back as a normal response with the stop reason "refusal", the documentation says. Developers can turn on a default fallback that retries the request on the model Anthropic recommends for that category.
Outside evaluators, including Frontier Design and METR, tested Opus 5.5 before release, Anthropic said and The Next Web reported. On its automated behavioral audit, the broadest alignment test it runs, Anthropic called Opus 5.5 "the strongest-performing model we've tested to date." In a new test of whether a model tries to cross containment boundaries, it "attempted to circumvent boundaries around 85% less often than Opus 5 or Claude Mythos 5.1," the company wrote, and it is more resistant than Opus 5 to prompt injection, where hidden instructions in a document or web page try to steer an agent.
Two programmes for verified professionals
For security teams, Anthropic says it will "soon" expand its Cyber Verification Program to cover Opus 5.5. "The new program will include three tiers for increasingly permissive trusted access, including access to Claude Mythos models."
Biology research runs through the Life Sciences Verification Program, which Anthropic launched in beta on 17 September. For now it takes teams and institutions, not individual plans, and reviews research credentials, security standards and ethical oversight. Standard Use grants cover a whole team and renew every year; High-risk Use grants are a project specific add-on renewed every six months. At launch the programme covered Mythos 5.1, Opus 5 and Sonnet 5 plus later models, and high risk grants for Mythos stayed limited to a small set of entities while Anthropic works with the US government. In the Opus 5.5 announcement the company said the programme gives "vetted organizations like academic labs, startups, and pharmaceutical companies access to safeguards designed for the full breadth of biology-related work."
What changes for developers
Opus 5.5 is available on the Claude API and on Amazon Web Services, Google Cloud and Microsoft's Foundry, as claude-opus-5-5 (anthropic.claude-opus-5-5 on Amazon Bedrock). It has a one million token context window and up to 128,000 output tokens, Anthropic's model page says, as MarkTechPost also reported. Four changes can break code written for Opus 5: thinking can no longer be switched off, forced tool use returns an error, thinking blocks are tied to the model and the conversation so that edits to earlier turns invalidate them, and the Claude API and Google Cloud reject the older computer_20251124 computer use tool. The default effort level drops to medium, from high on Opus 5.
For subscribers, Anthropic raised five hour usage limits on Pro, Max, Team and seat based Enterprise plans and added a rate limit reset that users can save for later. As with Fable 5.1, outputs carry the watermark Anthropic adds for the EU AI Act.
What happened next
Anthropic said Claude Sonnet 5.5 and Claude Haiku 5.5 would follow "in the coming weeks". Sonnet 5.5 arrived on 28 September at Sonnet 5's price of USD 2 and USD 10 per million tokens, with cache reads at USD 0.20. Our report on Claude Fable 5.1 covers the model Opus 5.5 is measured against.
Comments
Comments are reviewed before they appear.
No comments yet. Be the first to share your thoughts.