Anthropic released Claude Opus 4.8 on May 28, 2026 — 41 days after 4.7, the fastest upgrade cycle the company has shipped for Opus. The new model maintains identical pricing to its predecessor while adding two headline features: Dynamic Workflows, a research preview for orchestrating large parallel subagent pipelines, and Effort Controls, which let users trade compute depth for speed.
Dynamic Workflows
Dynamic Workflows launches in research preview inside Claude Code for Enterprise, Team, and Max plan subscribers. The feature lets a single session orchestrate tens-to-hundreds of parallel subagents, coordinate them across independent task angles, deploy adversarial agents to stress-test conclusions, and iterate with resumable state until answers converge before reporting back.
The stated use case is codebase-scale work: Anthropic says Opus 4.8 with Dynamic Workflows can carry a migration across hundreds of thousands of lines from kickoff to merge, using the existing test suite as its quality bar. For engineers running large refactors or platform upgrades, that's a meaningful scope expansion beyond what a single linear context window can hold.
Effort Controls
Effort Controls is available on claude.ai and Cowork across all subscription tiers. Turning effort up enables deeper extended thinking; turning it down prioritises speed and preserves rate-limit headroom. The feature surfaces an existing model capability as a user-facing dial rather than a system-prompt parameter.
The benchmark numbers
Opus 4.8's biggest gains are in agentic coding, computer use, and knowledge work. Honesty metrics are the other standout: the model scores 0% on uncritically reporting flawed results — the first Claude model to do so — and shows a ten-fold reduction in overconfidence versus Opus 4.7.
Coding
| Benchmark | Opus 4.8 | Opus 4.7 | GPT-5.5 |
|---|---|---|---|
| Terminal-Bench 2.1 | 88.5% | — | — |
| SWE-bench Pro (agentic coding) | 69.2% | 64.3% | 58.6% |
Agents & tool use
| Benchmark | Opus 4.8 | Opus 4.7 | GPT-5.5 |
|---|---|---|---|
| OSWorld-Verified (computer use) | 84% | 78.0% | — |
| Online-Mind2Web (browser agents) | 84% | — | — |
| GDPval-AA (knowledge work) | 1890 | 1753 | 1769 |
| HLE with tools (reasoning) | 57.9% | 54.7% | — |
Legal
Opus 4.8 is the first model to break 10% on the Legal Agent Benchmark's all-pass standard — a pass/fail eval that requires every clause in a legal task to be handled correctly, not just most of them. Prior models cleared individual tasks while failing on the end-to-end bar.
Honesty and reliability
| Metric | Opus 4.8 | Opus 4.7 |
|---|---|---|
| Uncritically reports flawed results | 0% | >0% |
| Fails to flag important events to user | 3.7% | — |
| Overconfidence rate | ~10× lower | baseline |
Early testers also report Opus 4.8 is four times less likely than its predecessor to let code flaws pass without flagging them.
Pricing and availability
Regular pricing is unchanged from 4.7: $5 per million input tokens and $25 per million output tokens. Fast mode — where the model runs at 2.5× the speed — is now available at $10/$50 per million tokens, which Anthropic says is three times cheaper than fast mode was for prior Opus releases.
The model is live on claude.ai, the Anthropic API (claude-opus-4-8), Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
Why it matters for working professionals
For lawyers, the Legal Agent Benchmark result is notable: prior models cleared most tasks but fell apart on all-pass standards, which more closely mirror real document work where a single missed clause can matter. For engineers, Dynamic Workflows removes a practical ceiling on what a single agentic session can accomplish. For anyone using Claude Projects on a Pro or Team plan, the upgrade is automatic — no migration required.
The honesty improvements are the quieter gain. A model that flags its own uncertainty rather than papering over it is worth more in professional contexts where unchecked errors compound quickly.
Sources: Anthropic · TechCrunch · The New Stack · SiliconANGLE