Claude Opus 4.8 for Coding: The Default for Serious Agentic Work

Claude Opus 4.8 was released on May 28, 2026, and since Fable 5 arrived above it, its role has actually gotten clearer, not smaller. Per Anthropic's own guidance, Opus is the recommended starting point for complex agentic coding: effort defaults to high, it is thorough on multi-file changes, and it carries a 1M tokens context window. The API identifier is claude-opus-4-8.

The economics are the story. Opus 4.8 delivers most of Fable 5's capability at exactly half the price, which makes it the value flagship of the lineup, the model you default to and only move away from with a reason. This guide covers the verified numbers, then the two reasons to move: escalating up to Fable 5 or dropping down to Sonnet 5. For the full lineup, see the best Claude model for coding.

Data verified July 18, 2026 against Anthropic's published pricing and docs. Note the symmetry: Opus 4.8's fast mode costs $10 in / $50 out per MTok, exactly Fable 5's standard rate. At that price point you are choosing between a faster Opus and a smarter Fable, and for latency-sensitive interactive sessions, faster Opus is often the right call.

Opus 4.8 posts the second-best Claude scores on the board, behind only Fable 5. The verified numbers:

Data verified July 18, 2026. Read the gap honestly: Fable 5 is measurably ahead, 95.0% vs 88.6% on Verified and 80.3% vs 69.2% on Pro, but Opus earns those scores at half the cost per token. For most sprints that trade is the whole argument. How it stacks against OpenAI's flagship is covered in Claude Opus 4.8 vs GPT-5.6 Sol.

Fable 5 costs $10 in / $50 out per MTok, double Opus, and there are three task shapes where it earns that. First, hard cross-cutting work: architecture-level refactors that touch a codebase everywhere, where the benchmark gap shows up in the diff. Second, long autonomous runs, where a wrong turn at hour two costs more than the token premium ever will. Third, spec drafting and review on large repos, where sustained reasoning over a full 1M window matters.

In the other direction, Sonnet 5 costs $3 in / $15 out per MTok, and intro pricing runs $2 in / $10 out per MTok through August 31, 2026, a fraction of Opus. Drop down for routine, well-specified work: standard feature stories, test suites, mechanical refactors with a clear plan, and high-volume parallel fleets where cost dominates. Sonnet is noticeably behind Opus on hard multi-step reasoning, so the rule is simple: if the story needs judgment, keep Opus; if it needs throughput, Sonnet is the rational pick.

The full three-tier decision, including where Haiku 4.5 fits below Sonnet, is mapped in Opus vs Sonnet vs Haiku.

Claude Code ships an alias that encodes the routing logic above into a single setting:

/model opusplan

With opusplan, Claude Code uses Opus for plan mode and Sonnet for implementation. The reasoning-heavy step, understanding the codebase and deciding what to change, gets the flagship; executing the approved plan runs on Sonnet's pricing. For teams doing spec-driven development, where the plan is reviewed before implementation starts, it is the cheapest way to keep Opus-quality judgment in the loop without paying Opus rates for every generated line.