Claude Opus 4.8 Ships With Dynamic Workflows — Same $5/$25 API Price

Anthropic released Claude Opus 4.8 on May 28, 2026 — just 41 days after the controversial Opus 4.7 launch. The API model ID is claude-opus-4-8. Pricing is unchanged at $5 per million input tokens and $25 per million output tokens. The headline upgrade for developers is Dynamic Workflows in Claude Code: parallel subagents aimed at codebase-scale migrations across hundreds of thousands of lines.

What Shipped in Claude Opus 4.8?

Opus 4.8 is Anthropic's frontier model refresh, available on Claude.ai, Claude Code, the Claude API, Amazon Bedrock, Google Vertex AI, and Microsoft Foundry the same day.

Core product changes beyond raw benchmark gains:

Dynamic Workflows (research preview) — Claude Code can orchestrate large problems across hundreds of parallel subagents. Anthropic's positioning: full repository migrations from kickoff to merge with the existing test suite as the bar.

Effort controls on Claude.ai — users set how hard Claude works on a task (speed vs depth tradeoff without switching models manually).

Fast mode pricing cut — Opus fast mode runs at roughly 2.5× speed at $10/$50 per million tokens (input/output), described as three times cheaper than prior fast-mode pricing for equivalent models.

1M context variant — developers can target claude-opus-4-8[1m] for long-context workloads.

What Do the Benchmarks Say?

Third-party summaries of Anthropic's published results highlight ~69.2% on SWE-Bench Pro for Opus 4.8 — a meaningful coding lift over 4.7.

The honest caveat from early analyses: Opus 4.8 does not win every terminal. Some comparisons show GPT-5.5 ahead on Terminal-Bench 2.1 (~78% vs ~75% in one cited breakdown). For agentic coding teams, the right move is task-specific evals, not model religion.

For enterprise context after Anthropic's $965B Series H round, see Anthropic Hits $965B Valuation After $65B Series H. For API cost comparisons, use the LLM API Pricing Tracker.

What Developers Should Change Today

Re-point default Claude Code models if 4.7 frustration (arguing, instruction drift) blocked adoption. Early user narrative positions 4.8 as a behavior correction release, not just incremental intelligence.

Pilot Dynamic Workflows on one migration — pick a repo with a real test harness. The value claim is end-to-end migration; verify on your codebase before betting a release train.

Keep fast mode for latency-sensitive loops — interactive refactors and CI-adjacent bots benefit from 2.5× speed if budget allows the $10/$50 tier.

Log model ID explicitly — pin claude-opus-4-8 in production configs; 4.7 sunset timelines will matter for regression tests.

Key Takeaways

• May 28, 2026: Claude Opus 4.8 live on API, Claude.ai, Claude Code — model ID claude-opus-4-8
• Pricing unchanged: $5/$25 per million tokens standard; fast mode $10/$50 at ~2.5× speed
• Dynamic Workflows: parallel subagents in Claude Code for large-scale codebase work (research preview)
• SWE-Bench Pro ~69% cited for 4.8 — verify against your tasks; not universal terminal leadership
• For developers: upgrade defaults if 4.7 blocked you; pilot workflows on one migration with real tests
• What to watch: Dynamic Workflows GA timeline; 4.7 deprecation dates; Bedrock/Vertex rollout parity

Sources

• Introducing Claude Opus 4.8 — Anthropic, May 28 2026
• Anthropic releases Opus 4.8 with dynamic workflow tool — TechCrunch
• Claude Opus 4.8: what changed — benchmark summary