Claude Opus 4.8

Anthropic's Claude Opus 4.8 shipped on 28 May 2026 with a 1M token context window, priced at $5 per million input tokens and $25 per million output. Artificial Analysis ranks it second overall on GDPval-AA at 1615 Elo, behind Claude Fable 5 (1783) and ahead of GLM-5.2 (1524) and GPT-5.5 at xhigh (1509). The long-context claim carries a number attached to it: 68.1% F1 on GraphWalks at the full 1M tokens, an improvement over Opus 4.7.

That window is also where the practical critique lands. Writing on 29 May 2026, the day after release, Pasquale Pillitteri argued that most SEO users waste the model inside a single giant conversation, mixing strategy, keyword research and drafting until the context drifts toward generic best practices. A million tokens makes retrieval easy and does nothing to separate a strategic decision from an operational brief. His routing sends Opus 4.8 to strategy and pillar pages, Sonnet 4.6 to production and Haiku 4.5 to micro-tasks, grounding each step in live data through MCP servers, for $1 to $3 per article.

Price positioning matters in the same way. Didier Girard cites Fable 5 at $10/$50, twice Opus 4.8's rate, while open-weights models push from below. On Z.ai's private Code Bench, published 14 August 2026, GLM-5.3 scores 31.4% at roughly 50K output tokens against 29.5% at 120K for Claude Opus 4.8. Nearly all of those evaluations were run inside Claude Code 2.1.207.