Introducing Claude Opus 4.8
Key Points
- 2.5× fast mode with lower fast-tier cost
- Dynamic workflows for codebase-scale parallel agents
- Effort control + Messages API for mid-run updates
Summary
Claude Opus 4.8 is available today with broad capability and reliability improvements over Opus 4.7. It ships with new developer-facing features (dynamic workflows, effort control, Messages API change), faster operation modes, and the same baseline pricing. The model shows measurable gains on coding, agentic tasks, reasoning, legal and web-agent benchmarks, and is reported to be more honest and better at flagging uncertainties.
Key Points
-
Availability and pricing
- Opus 4.8 is available everywhere today via the Claude API (model id: claude-opus-4-8).
- Regular pricing unchanged: $5 /M input tokens, $25 /M output tokens.
- Fast mode: 2.5× speedup and billed at $10 /M input, $50 /M output (fast mode is ~3× cheaper than previous-model fast tiers).
-
New features for engineering workflows
- Dynamic workflows (research preview in Claude Code for Enterprise, Team, and Max): plans + run hundreds of parallel subagents, verify outputs, scale to codebase-level migrations and long-running orchestrations.
- Effort control (claude.ai and Cowork; available on all plans): choose effort level (default: high; options include extra/xhigh and max) to trade token use for solution quality—use extra/xhigh for difficult or long-running async jobs.
- Messages API now accepts system entries inside the messages array to update instructions, permissions, token budgets, or environment context mid-task without breaking prompt cache.
-
Behavior & benchmarks
- Stronger agentic performance, improved tool-calling efficiency, higher-quality analysis and reasoning across long sessions.
- Notable wins: best-in-class on several internal benchmarks (legal, online web-agent) and measured reductions in unsupported claims (~4× less likely to let code flaws pass unremarked).
- Alignment assessment shows reduced misaligned behaviors and improved prosocial traits; see the Claude Opus 4.8 System Card for full details.
Action items for engineers
- Try the new effort control to tune latency, token use, and result quality for your workloads; prefer extra/xhigh for complex code tasks.
- Evaluate dynamic workflows for large-scale code migrations or massively parallel agent tasks (requires Enterprise/Team/Max access and is in research preview).
- Use the Messages API system entries to update runtime instructions (permissions, token budgets, environment) without a user-turn workaround.
- Update rate-limit and token-usage assumptions for Claude Code (higher limits available to accommodate higher-effort runs).
Links and references
- Model: claude-opus-4-8 via the Claude API
- See the Claude Opus 4.8 System Card and product post for full evaluations and alignment testing.