Anthropic’s Claude Opus 4.8 cuts fast mode costs by 3X, boosts agents

Anthropic has unveiled Claude Opus 4.8, the latest iteration of its premium AI model, introducing pricing parity with its predecessor alongside a significant reduction in "fast mode" costs. The update also introduces dynamic workflows that enable the model to deploy hundreds of parallel subagents for large-scale tasks, marking a step forward in agentic reasoning and efficiency.

The model rolls out immediately across all Anthropic platforms—including claude.ai, Claude Code, the API, and Cowork—at the same pricing as Opus 4.7: $5 per million input tokens and $25 per million output tokens. Developers can access it via the API using the identifier claude-opus-4-8.

Fast mode pricing drops 3X, unlocking high-throughput use cases

The most notable change is the overhaul of fast mode, which now costs $10 per million input tokens and $50 per million output tokens, down from $30 and $150 respectively in Opus 4.7. This 3X reduction makes high-speed inference significantly more affordable, bringing it within reach for latency-sensitive production workloads.

Fast mode is already available in Claude Code via the /fast command, while API access remains restricted and requires joining a waitlist. Industry observers note that this pricing shift could reshape cost structures for AI-driven automation, particularly in workflows where response time is critical.

Benchmark improvements and Mythos-class positioning

While Opus 4.8 delivers modest but meaningful gains over its predecessor, its performance remains competitive with leading models. On SWE-bench Verified, it scores 88.6% (up from 87.6% in Opus 4.7), while SWE-bench Pro improves to 69.2% (from 64.3%). Terminal-Bench 2.1 also sees a jump to 74.6% (from 66.1%).

Anthropic frames the model as a tangible step forward, outperforming GPT-5.5 in 12 benchmarks, including coding, agentic tool use, and long-context tasks. However, GPT-5.5 retains an edge in terminal/CLI workflows and graduate-level science tasks.

A key highlight is its placement between Opus 4.7 and the more advanced Claude Mythos Preview, which is currently restricted to select organizations under Project Glasswing for cybersecurity applications. Anthropic has hinted that Mythos-class capabilities will soon be available to all customers once additional safeguards are finalized.

Enterprise adoption and dynamic workflows

Early adopters report tangible benefits. Databricks noted a "step change in agentic reasoning" within its Genie data agent, achieving 61% lower token costs than Opus 4.7 due to improved multimodal efficiency on PDFs and diagrams. Hebbia, a financial analytics firm, praised its superior citation precision and token efficiency when processing dense financial filings.

Cognition, the maker of the Devin AI coding agent, observed that Opus 4.8 addresses comment-verbosity and tool-calling issues present in Opus 4.7, translating to faster engineering productivity. A third-party vendor also reported an 84% score on Online-Mind2Web, surpassing both Opus 4.7 and GPT-5.5.

The update also introduces dynamic workflows in Claude Code, a feature designed for large-scale tasks that exceed a single context window. The model autonomously plans, deploys hundreds of parallel subagents, executes the task, and validates outputs before delivering results. Anthropic cites a use case involving a multi-thousand-line codebase migration, where the system handled the entire process from kickoff to merge while relying on existing test suites for validation.

This feature is available on Claude Code’s Enterprise, Team, and Max plans.

New controls for efficiency and API flexibility

Anthropic has added two smaller but practical features:

• Effort control in claude.ai and Claude Cowork: Users can now select how much computational effort the model expends per response. Higher effort modes consume more tokens for refined answers, while lower effort modes prioritize speed and token efficiency.

• Mid-task system entries in the API: Developers can now update Claude’s instructions dynamically during an active session, adjusting permissions, token budgets, or environmental context without restarting the task.

With these updates, Anthropic continues to refine its flagship model, balancing cost efficiency, performance, and scalability for both developers and enterprises. The shift in fast mode pricing, alongside enhanced agentic capabilities, signals a move toward making high-performance AI more accessible without compromising quality. As the company prepares to expand Mythos-class features to a broader audience, the industry watches closely to see how these advancements will influence the next wave of AI-driven workflows.