Anthropic launched Claude Opus 4.8 on May 28, 2026, equipping the upgraded model with an explicit effort dial that lets users trade faster, lower-cost responses against deeper reasoning. The release, available immediately through claude.ai, the API and cloud partners, also brought a revamped fast mode priced at $10 per million input tokens and $50 per million output tokens—one-third the cost of its predecessor’s fast tier, according to the company.
The new controls reflect a broader push among frontier labs to offer tunable inference economics rather than betting everything on headline benchmark numbers. Other labs have introduced similar compute‑budget controls, but Anthropic is making the choice explicit next to the model selector. The company is not yet delivering a generational leap—its more advanced Mythos model remains in preview—but by turning reasoning depth into a product feature, it is responding to enterprise demand for predictable cost and latency tradeoffs.
The effort control appears alongside the model selector in claude.ai and the Cowork assistant, with “high” set as the default. Users can choose lower effort for faster replies that consume rate limits more slowly, or step up to “extra” (labeled “xhigh” in Claude Code) and “max,” which cause the model to think more frequently and deeply. The highest setting significantly increases token usage, though Anthropic has not disclosed the exact consumption mapping.
Standard API pricing remains unchanged from Opus 4.7: $5 per million input tokens and $25 per million output tokens. The restyled fast mode runs at up to 2.5 times the speed of regular usage and, at $10 input and $50 output per million tokens, is still a premium over standard but, Anthropic says, is three times cheaper than the fast mode that accompanied previous models. The unchanged base price and cheaper fast tier signal an effort to make frontier AI more accessible for production workloads, not just experiments. The fast mode is in research preview, meaning availability and quotas could shift.
Anthropic says Opus 4.8 improves on its predecessor across coding, agentic tasks, reasoning and knowledge work, calling it the company’s most capable generally available model in those domains. In one internal evaluation, the new model was roughly four times less likely to let unremarked flaws slip into its own code. However, most benchmark data are company-reported, and independent replication is thin. Detailed safety and capability test results were promised in a system card, but that document was not independently reviewed ahead of publication.
For developers, Claude Code gained a research preview of dynamic workflows that lets the model orchestrate hundreds of parallel subagents in a single session, plan complex workstreams and verify outputs. The feature is aimed at large-scale codebase migrations and other high-stakes engineering tasks, but its stability and cost at scale remain unproven. On the same day, AWS confirmed the model is available on Amazon Bedrock and the Claude Platform on AWS, and GitHub said Opus 4.8 is generally available in Copilot.
The model announcement arrived alongside a separate $65 billion Series H fundraise at a $965 billion post-money valuation, which Anthropic framed as a distinct financing event. The product story stands apart: with a user-facing effort dial, repriced fast mode and new developer workflows, the company is giving operators more handles on the cost-quality curve. How accurately those handles translate into real-world performance and savings will depend on independent testing that has yet to materialize.