What's the difference between Claude Opus 4.7 and 4.8?

Three things. First, Opus 4.8 has Dynamic Workflows in Claude Code that let it dynamically write JavaScript orchestration scripts to spin up tens to hundreds of parallel subagents in a single session — the demo migrated a 750,000-line codebase from Zig to Rust in 11 days with 99.8% test pass rate. Second, Fast Mode runs roughly 2.5x faster and is roughly 3x cheaper than Opus 4.7 Fast, dramatically improving the cost curve for high-volume agentic workloads. Third, Opus 4.8 scores higher on agentic benchmarks (SWE-Bench Pro 69.2% vs Opus 4.7 Adaptive's 64.3%) and is described as more honest about flagging uncertainty rather than hallucinating. API pricing for standard mode is unchanged from Opus 4.7.

Is Claude Opus 4.8 better than GPT-5.5 for coding?

It depends on the workload. Opus 4.8 leads on SWE-Bench Pro (69.2% vs GPT-5.5's 58.6%) and is described as more reliable for complex multi-file refactors, code review, and unattended agent workloads. GPT-5.5 wins on terminal-coding benchmarks and on OpenAI's internal Expert-SWE test (73.1% on roughly 20-hour tasks), making it strong for long-horizon CI-style fixers. Many production teams now run a router: Opus 4.8 for high-stakes reasoning and code review, GPT-5.5 for token-sensitive multi-step terminal automation, Gemini 3.5 Flash for high-volume cheap tasks. Pick by job, not by overall ranking.

Quick Answer

What Is Claude Opus 4.8? Anthropic May 28, 2026 Release Guide

Q: What is Claude Opus 4.8?

Claude Opus 4.8 is Anthropic's flagship model released on May 28, 2026, replacing Opus 4.7 as the default top-tier model on the Claude API, Claude.ai, AWS Bedrock, Google Vertex, and Microsoft Foundry. It introduces three major shipping features: Dynamic Workflows (parallel subagent orchestration in Claude Code, capped at 1,000 agents per run with 16 concurrent), a Fast Mode that is roughly 2.5x faster and 3x cheaper than Opus 4.7 Fast, and improved agentic computer use plus financial reasoning. On Anthropic's internal SWE-Bench Pro benchmark it scores 69.2%, beating GPT-5.5 and Gemini 3.1 Pro on most evaluations though GPT-5.5 leads on terminal coding.

Q: How do I use Claude Opus 4.8?

Three paths. (1) Claude.ai chat — sign in to claude.ai, Opus 4.8 is the default for Pro, Max 5x, and Max 20x subscribers. (2) Claude Code CLI — install with 'npm i -g @anthropic-ai/claude-code', set ANTHROPIC_API_KEY, and run 'claude' in your project. Dynamic Workflows trigger automatically for tasks the model judges large enough. (3) API — use the model ID 'claude-opus-4-8-20260528' against the Anthropic API, AWS Bedrock, Google Vertex, or Microsoft Foundry. Most existing Opus 4.7 code works unchanged — just swap the model ID.

Published: May 31, 2026

What Is Claude Opus 4.8? Anthropic May 28, 2026 Release Guide

Claude Opus 4.8 is Anthropic’s flagship model released May 28, 2026 — replacing Opus 4.7 across Claude.ai, the Anthropic API, AWS Bedrock, Google Vertex, and Microsoft Foundry. Three big shipping features: Dynamic Workflows (parallel subagents in Claude Code), a much cheaper Fast Mode (~3x lower cost, ~2.5x faster), and stronger agentic reasoning (SWE-Bench Pro 69.2%).

Last verified: May 31, 2026.

TL;DR

Feature	Status
Release date	May 28, 2026
Model ID	`claude-opus-4-8-20260528`
Standard pricing	~$15/M input, ~$75/M output (unchanged from Opus 4.7)
Fast Mode	~2.5x faster, ~3x cheaper than Opus 4.7 Fast
Context window	200K tokens
SWE-Bench Pro	69.2% (beats GPT-5.5 58.6%, Gemini 3.1 Pro lower)
Headline feature	Dynamic Workflows in Claude Code
Availability	API, Claude.ai (Pro / Max), Bedrock, Vertex, Foundry

The headline: Dynamic Workflows

This is the biggest change in agentic AI in May 2026. Inside Claude Code, when Opus 4.8 judges a task large enough, it now writes a JavaScript orchestration script that spawns tens to hundreds of parallel subagents — capped at 1,000 total agents per run with 16 concurrent.

The architectural shift: orchestration logic lives in code, not in Claude’s context window. Claude generates the JS script, a runtime executes it in the background, intermediate results are managed by the script, and only the final aggregated answer comes back into Claude’s context. Adversarial verifier agents review subagent output before integration.

Reference demo from Anthropic: Zig → Rust migration of a 750,000-line codebase, completed in 11 days with 99.8% test pass rate. Real-world early reports show 5–20x speedups on:

Codebase-scale migrations (framework upgrades, language ports)
Security audits across hundreds of files
Bug-hunt sweeps over an entire repo
Refactors that touch many call sites

Dynamic Workflows are available on all paid Claude Code plans (Pro, Max 5x, Max 20x, Team, Enterprise) via CLI, Desktop, and the VS Code extension, plus the API / Bedrock / Vertex / Foundry. Note: they burn significantly more tokens than normal Claude Code sessions — pilot on a scoped task before letting it loose on production work.

Fast Mode: ~3x cheaper, ~2.5x faster

Opus 4.8 Fast Mode (Anthropic’s lower-latency variant for non-extended-thinking traffic) runs roughly 2.5x faster and at roughly one-third the cost of Opus 4.7 Fast. Standard mode pricing is unchanged. This makes Opus 4.8 a credible choice for high-volume agentic workloads that would previously have routed to Sonnet 4.6 or Haiku — closing the gap with Gemini 3.5 Flash and GPT-5.5 on cost-per-task for medium-difficulty work.

Benchmarks vs the field

Internal Anthropic benchmarks plus public scoreboards (Scale Labs, BenchLM, LLM-Stats) as of May 30, 2026:

Benchmark	Opus 4.8	GPT-5.5	Gemini 3.5 Flash
SWE-Bench Pro	69.2%	58.6%	lower
OpenAI Expert-SWE (~20h tasks)	competitive	73.1%	lower
Coding average (BenchLM aggregate)	76.4	competitive	54.5
HLE	57.9%	competitive	40.2
Terminal coding	strong	leads	weaker
Multimodal/grounded	competitive	competitive	leads

Claude Mythos Preview (Anthropic’s next-gen model, not yet GA) sits at 77.8% SWE-Bench Pro but is invite-only research preview.

What’s actually new beyond the headline

Beyond Dynamic Workflows and Fast Mode, Anthropic shipped a cluster of features alongside Opus 4.8:

Self-hosted Sandboxes (Public Beta) — tool execution runs in your infrastructure (or a managed provider), agent orchestration stays on Anthropic’s. Critical for regulated industries.
MCP Tunnels (Research Preview) — agents access MCP servers inside a private network without exposing them to the public internet.
Effort Control — claude.ai and Claude Cowork users can dial how much effort Claude puts into a response, trading speed for depth.
Doubled 5-hour rate limits across paid Claude Code plans.
Considerably raised Opus API rate limits, backed by the SpaceX/Colossus compute partnership announced earlier in May.
Microsoft 365 add-ins for Excel, PowerPoint, Word (GA) and Outlook (public beta).

How to use Claude Opus 4.8

Path 1: Claude.ai chat

Sign in at claude.ai. Opus 4.8 is the default top-tier model for Pro ($20/mo), Max 5x ($100/mo), and Max 20x ($200/mo) subscribers. Just start chatting.

Path 2: Claude Code CLI

npm i -g @anthropic-ai/claude-code
export ANTHROPIC_API_KEY=sk-ant-...
cd your-project
claude

Dynamic Workflows trigger automatically for tasks Opus 4.8 judges large enough (codebase-scale migrations, audits, bulk refactors). For smaller tasks, it behaves like Opus 4.7 — fast, focused, single-threaded.

Heads up: the June 15, 2026 Claude Code billing change applies. Programmatic usage moves to a separate credit pool ($20 / $100 / $200 by tier, metered at full API rates, no rollover). For heavy use, switch to direct API billing.

Path 3: API

import anthropic
client = anthropic.Anthropic()
message = client.messages.create(
    model="claude-opus-4-8-20260528",
    max_tokens=4096,
    messages=[{"role": "user", "content": "..."}]
)

Same shape as Opus 4.7 calls — just change the model ID. Bedrock and Vertex publish their own model IDs; check the respective consoles. Microsoft Foundry exposes Opus 4.8 in Copilot Studio’s multi-model picker.

Who should switch

Situation	Switch to Opus 4.8?
Heavy Claude Code user	Yes — Dynamic Workflows alone justify it for any non-trivial refactor work.
Production agent stack on Opus 4.7	Yes — same standard pricing, better benchmarks, drop-in.
High-volume cheap workloads	Maybe — Fast Mode cost cut closes the gap with Gemini 3.5 Flash; still pricier per token.
Terminal-CI fixers	No — GPT-5.5 still leads here.
Multimodal-heavy / massive context	No — Gemini 3.5 Flash 1M context still wins for that profile.
Stuck on Opus 4.6 / Opus 4.7 budget tier	Yes — Opus 4.7 will deprecate; migrate before the timeline lands.

Verdict

Opus 4.8 is a real step forward, not a point release. Dynamic Workflows make Claude Code meaningfully better at codebase-scale work; the Fast Mode cost cut makes Opus 4.8 viable for use cases that previously routed to Sonnet; agentic benchmarks moved up materially. For anyone running Claude Code or agentic backends today, the upgrade is essentially free — same standard pricing, better model, new features. Mythos Preview is the next frontier but is not yet GA — Opus 4.8 is what production should be on right now.

Sources: Anthropic Opus 4.8 launch post (May 28, 2026), Marktechpost Dynamic Workflows analysis, BenchLM SWE-Bench Pro leaderboard, Scale Labs SWE-Bench Pro public leaderboard, Microsoft Foundry release notes, Anthropic API pricing docs (verified May 31, 2026).