The capable everyday model
Claude Opus 4.8 is a high-capability model Anthropic released in 2026. It belongs to the Opus tier and is aimed at coding, AI agents, reasoning, and professional knowledge work.
It is a hybrid reasoning model handling context up to one million tokens, intended for sustained multi-step work across long documents and large codebases.
What improved
Anthropic describes better judgment in coding and agentic work, better tool use, and more stability over long-running tasks than the previous generation. It is also more inclined to push back when a premise looks wrong, and to flag the parts of its own output it is least sure about.
In Claude Code it can carry investigation, design, implementation, testing, and repair as one continuous piece of work. It is equally usable for writing, research, and document preparation.
How it differs from Fable 5
Fable 5 is the Mythos-class top model for especially hard or long, complex problems. Opus 4.8 sits below it as the capable everyday choice, balancing quality, speed, and usage allowance.
Sending planning and difficult design to Fable 5 while keeping routine implementation and writing on Opus 4.8 works well. Choosing by difficulty beats defaulting to the top model for everything.
Getting the most from it
Filling a long context with material does not mean the important parts get organized for you. Pairing it with context engineering — stating the goal, the priorities, and which sources matter — is what makes the capability usable.
Even a top-tier model needs its code checked for safety and its claims checked for accuracy. Do not skip testing and human review.