Skip to content
News Model release

Anthropic ships Claude Opus 4.8 with dynamic workflows and cheaper fast mode

· by Pondero Newsdesk

The short version

Anthropic released Claude Opus 4.8 on May 28, 2026, pairing the model upgrade with a new dynamic workflows feature for Claude Code and a fast-mode price cut of roughly three times compared to the previous generation.

Anthropic ships Claude Opus 4.8 with dynamic workflows and cheaper fast mode

Anthropic released Claude Opus 4.8 on May 28, 2026, paired with a new agentic feature called dynamic workflows and a meaningful cut to fast-mode pricing. The model is available immediately through the Claude API, claude.ai, and GitHub Copilot.

What

Opus 4.8 replaces Opus 4.7 at the same base price: $5 per million input tokens and $25 per million output tokens for regular mode (per Anthropic). Fast mode, which runs the model at 2.5 times the standard speed, drops to $10 per million input tokens and $50 per million output tokens. Anthropic described that as roughly three times cheaper than fast mode was on previous Opus models (per Anthropic).

On benchmarks, early testers reported substantive gains in coding and agentic work. Cursor co-founder and CEO Michael Truell said Opus 4.8 exceeded prior Opus models at every effort level on CursorBench, and that tool calling became more efficient, using fewer steps to reach the same result (per Anthropic). Devin CEO Scott Wu said the model uses tools more cleanly and follows instructions with greater consistency than Opus 4.7, and that it fixed comment-verbosity and tool-calling problems his team saw with the prior release (per Anthropic). CoCounsel Legal head of applied research Niko Grupen confirmed Opus 4.8 is the first model to break 10 percent on the Legal Agent Benchmark all-pass standard (per Anthropic). One tester, identified as working on browser agent software, cited 84 percent on Online-Mind2Web (per Anthropic).

Anthropic also highlighted a honesty-related improvement: Opus 4.8 is around four times less likely than its predecessor to let flaws in code it has written pass unremarked (per Anthropic). The model is also more likely to flag uncertainties in agentic loops rather than assert false progress.

Dynamic workflows

Alongside the model, Anthropic launched dynamic workflows for Claude Code in research preview. The feature applies to Enterprise, Team, and Max plan users. On the API, it is available on Amazon Bedrock, Vertex AI, and Microsoft Foundry as well (per Anthropic).

The mechanism: instead of running a single agent through a task sequentially, Claude writes its own orchestration scripts, fans the work out across tens to hundreds of parallel subagents in one session, checks each result, and returns a consolidated answer. Anthropic positioned it for problems too large for one pass: codebase-wide bug hunts, framework migrations that touch thousands of files, and high-stakes work where running multiple independent agents against the same problem produces more reliable output (per Anthropic).

A concrete example from the announcement: developer Jarred Sumner used dynamic workflows to port the Bun runtime from Zig to Rust, producing roughly 750,000 lines of Rust with 99.8 percent of the existing test suite passing, in eleven days from first commit to merge (per Anthropic). One workflow mapped Rust lifetimes for every struct in the Zig codebase. Subsequent workflows wrote each Rust file as a behavior-identical port, ran build-and-test loops until clean, and then opened pull requests for an optimization pass. All of this was handled by dynamic workflows before entering production review.

Klarna senior engineering manager Alessio Vallero described the feature as useful for dead-code identification across large codebases, a task where traditional static analysis had missed results (per Anthropic).

Anthropic was direct about the cost: dynamic workflows consume substantially more tokens than a typical Claude Code session. Users on Max and Team plans have it on by default. Enterprise users have it off by default, with an admin toggle to enable it. The first time a workflow triggers, Claude Code prompts for confirmation before proceeding.

GitHub Copilot availability

GitHub made Opus 4.8 generally available across its Copilot platform on the same day. The model is accessible to Copilot Pro+, Business, and Enterprise users across Visual Studio Code, Visual Studio, Copilot CLI, the GitHub Copilot cloud agent, GitHub Mobile, JetBrains, Xcode, and Eclipse (per GitHub). Rollout is gradual. Business and Enterprise plan administrators must enable the Opus 4.8 policy in Copilot settings before users can select it.

GitHub noted a 15x premium request multiplier for Opus 4.8 until usage-based billing launches on June 1, 2026 (per GitHub). After that date, the model will be priced under GitHub's standard usage-based billing framework.

Why it matters

The fast-mode price cut is the practical change that affects the most developers right now. Fast mode has been a throughput option for high-volume agentic pipelines, but at its previous price it was prohibitive in sustained production use. At roughly one-third the prior cost for the same 2.5x speed multiplier, it becomes viable for more routine tasks.

Dynamic workflows represent a structural shift in what an AI coding tool can take on autonomously. A single Claude Code session orchestrating hundreds of subagents, with progress saved so an interrupted job resumes rather than restarts, changes the size category of task a solo developer can delegate. The Bun port example is a meaningful reference point: 750,000 lines in eleven days is not a benchmark score; it is a production codebase change.

The trade-off is determinism. A model that writes its own orchestration plan and fans work across parallel agents is harder to debug and audit than one following a fixed sequence. Anthropic kept Enterprise users opted out by default, which signals the company's own read on organizational readiness.

The Anthropic alignment team's assessment of Opus 4.8, as reported in the announcement, found rates of misaligned behavior substantially lower than Opus 4.7, and comparable to Claude Mythos Preview. That comparison is notable because Mythos Preview is currently restricted to Project Glasswing cybersecurity partners while Anthropic develops additional safeguards for wider release.

Context and reactions

The Opus 4.8 launch arrived on the same day Anthropic announced a $65 billion Series H funding round at a $965 billion post-money valuation (per Anthropic news). Dynamic workflows is one of the first visible products reflecting the scale of investment now behind the platform.

Databricks CTO for neural networks Hanlin Tang said Opus 4.8 unlocks a step change in agentic reasoning in Genie, Databricks' AI agent for data and knowledge work, and cited 61 percent cheaper token cost than Opus 4.7 for multimodal workloads (per Anthropic). These are vendor-attributed productivity claims and have not been independently benchmarked.

Anthropic also announced that Mythos-class models are expected to reach all customers in the coming weeks, following progress on the cyber safeguards required for general availability (per Anthropic).

What to watch next

Three things are worth tracking. First, whether dynamic workflows move from research preview to general availability, and what pricing looks like under usage-based billing when token consumption at scale becomes clearer. Second, the Mythos-class release timeline: Anthropic said "coming weeks" as of May 28. Third, whether the Legal Agent Benchmark all-pass result attributed to CoCounsel (per Anthropic) is replicated by outside evaluators, given CoCounsel is a vendor early-tester and the claim is sourced from their head of applied research.

Sources