Anthropic just made its premium Claude model cheaper and faster without, according to the company, giving up flagship-level performance.
Claude Opus 5.5 is the first model in Anthropic’s new 5.5 generation, with lower token pricing, faster output, and benchmark results that the company says put it near or ahead of its existing top models on several coding and knowledge tests.
The launch comes as AI labs compete increasingly on economics as well as raw capability, with businesses paying close attention to how much useful work autonomous models can complete for every dollar spent.
Cheaper tokens, sprawling code
Anthropic set Opus 5.5 pricing at $4 per million input tokens and $20 per million output tokens, a 20% drop from Opus 5. Cache reads, which dominate costs during extended coding routines, dropped 60% to $0.20 per million tokens.
Those changes, according to Anthropic, can lower the cost of some typical workloads by around 40%, particularly in long-running tasks that repeatedly reuse cached context. The company also says the model generates output more than 30% faster.
Benchmark scores published by Anthropic place Opus 5.5 at 66.4% on Terminal-Bench 4.0, topping OpenAI’s GPT-6 Astra at 57.9%. Early testers reported significant gains on large codebases.
GitHub Chief Product Officer Mario Rodriguez said Opus 5.5 used among the fewest tokens and steps in GitHub’s testing across Copilot CLI and VS Code. In VS Code, he said, it solved more terminal tasks than Opus 5 while using less than half as many steps.
On knowledge evaluations, Opus 5.5 registered 1846 Elo on GDPval-AA v2.1 across 44 professions, outperforming Fable 5.1 and Opus 5.
Safeguards and evaluation limits
Frontier Design and METR evaluated Opus 5.5 before release. Anthropic said the model attempted to cross containment boundaries 85% less often than Opus 5 during behavioral testing.
Because its biological and cybersecurity capabilities rival Claude Mythos 5.1, Anthropic paired the launch with strict guardrails. General cybersecurity prompts automatically fall back to Opus 4.8. Access to higher-tier tasks requires vetted entry through Anthropic’s Cyber Verification Program or Life Sciences Verification Program.
One complication stood out: Anthropic said Opus 5.5 frequently recognized that it was being evaluated, making it harder to determine whether observed behavior would translate directly to real-world deployments.
Lower Model Prices Change the Enterprise Cost Equation
Lower pricing makes replacing older models with Opus 5.5 an obvious option to test, but model choice is only one part of an enterprise AI bill.
For long-running agentic workflows, using the most capable model for every step can still be expensive. Some organizations may get better economics by routing simpler tasks to lighter models while reserving higher-capability models for complex reasoning, coding, or recovery steps.
The trade-off is additional orchestration complexity. Multi-model systems require routing logic, evaluation, monitoring, and fallback behavior of their own, so the cheapest model mix on paper is not necessarily the cheapest system to operate.
More must-read AI coverage
- SS&C Intralinks DealCentre AI vs. Datasite: Which platform is built for the future of dealmaking?
- SS&C Intralinks FundCentre AI vs. Juniper Square: Which platform better supports modern private markets fund managers?
- Why Data, Not Models, Determines AI Success
- The Rise of the AI-Native Factory: How Physical AI Is Transforming Manufacturing
Operational impacts and user friction
For existing Claude developers, the update introduces notable friction alongside performance gains.
Anthropic made adaptive thinking mandatory, meaning engineers can no longer toggle the reasoning feature off; they can only calibrate depth via an effort parameter. The company also deprecated older tools, including the legacy computer-use interface, and tied thinking blocks directly to individual conversations to curb model distillation.
Teams with rigid API workflows face immediate engineering overhead to rewrite orchestration logic and update test suites. Subscribers on Pro, Max, Team, and Enterprise plans gain expanded five-hour usage limits and a storable rate-limit reset. Everyday users will also notice clearer prose, addressing persistent complaints regarding verbose, jargon-heavy outputs in previous releases.
Opus 5.5 is available immediately through Anthropic’s API, Amazon Web Services, Google Cloud, and Microsoft Azure, with Sonnet 5.5 and Haiku 5.5 due in the coming weeks.
For enterprises, the more important shift may be economic rather than benchmark-driven. As frontier models get cheaper and faster, the question is increasingly not simply which model is strongest, but how much useful autonomous work each model can deliver before cost and orchestration complexity erase the advantage.
Want to learn more AI tips, tricks, and prompting techniques? TechRepublic readers get free 7-day access to The Neuron Academy, our practical learning platform designed to help professionals use AI more confidently at work. Browse all lessons →
Let us teach you How to Talk to AI for free! Try our six-minute course at The Neuron Academy and learn a few simple ways to write better prompts and get more useful results from AI, or browse our other AI course for free for seven days. Check out all the lessons here →