Claude Opus 4.8

Opus 4.8 is the heavy-duty tier: $5 in, $25 out per million tokens, half of Fable 5 across the board. It is aimed squarely at engineering work — large codebases, multi-file changes, and review tasks where the model needs to hold a lot of interdependent context at once.

Where Sonnet is the model you use all day, Opus is the one you bring in when a change touches twenty files and getting it subtly wrong costs an afternoon of debugging.

Opus is at its best reading code it did not write. It follows call chains, notices when a change breaks an assumption three files away, and is willing to say a proposed approach is wrong rather than implementing it politely. For code review specifically, it flags edge cases and error paths that a human reviewer skimming a diff will miss.

It also handles genuinely long documents well — technical specifications, contracts, research papers — where the task is synthesis across the whole thing rather than retrieval of one fact.

It is slower than Sonnet by enough to notice in an interactive loop, and the price makes it wasteful for routine work. Opus writing boilerplate, renaming variables, or generating tests for a simple function is money burnt.

There is also a subtler cost: because it is careful, it produces longer, more hedged output. On a task that wanted a one-line answer, you pay for three paragraphs of reasoning you did not need.

Use Opus 4.8 for codebase-wide changes, serious code review, and analysis where the source material is long and interconnected. Drop to Sonnet 5 for day-to-day development — the gap is small on ordinary tasks and the cost difference is five-fold. If you are running agents, Opus for planning and Sonnet for execution is a pattern that works.