Trusted by users from 10,000+ companies
Take a quick look at new features and improvements that extend what teams can do with Opus 4.8 in production.
Run Opus 4.8 at up to 2.5x output speed via the API. Fast mode is priced at 3x less than it was for previous models, making speed-critical workflows more economically viable.
Available in Claude Code and enables planning and parallel execution of hundreds of subagents in a single session, with output verification before reporting back.
Users can now choose how much effort Claude puts into a response across all claude.ai plans. Higher effort means deeper thinking; lower effort means faster responses.
There is much more to discover about Claude Opus 4.8. Start by reading some of the user inquiries.
Developers can insert updated instructions mid-task via system entries in the messages array without breaking the prompt cache, reducing cost on long-running agentic loops.
Opus 4.8 responds directly on simple steps and thinking deeply on complex ones. Reduces wasted tokens on mixed workloads at the same effort level as Opus 4.7.
The minimum cacheable prompt length drops to 1,024 tokens on Opus 4.8. Prompts too short to cache on 4.7 now create cache entries with no code changes needed.
Opus 4.8 is less likely to skip a tool call a task required. An issue some users reported with 4.7. More reliable tool use across autonomous engineering and agentic workflows.
Long agentic traces stay on task with fewer derailments after compaction. Especially relevant for unattended, long-running workflows that previously stalled mid-execution.
Anthropic's alignment team found Opus 4.8 reaches new highs on supporting user autonomy and acting in the user's best interest, with misaligned behaviour rates comparable to Claude Mythos Preview.
Opus 4.8 leads on the benchmarks that reflect real professional and agentic work, not just test performance.
Claude Opus 4.8 delivers the highest score ever recorded on the Legal Agent Benchmark and is the first model to break 10% overall on the all-pass standard. The bar where every sub-task in a legal workflow is completed correctly.
On the Super-Agent Benchmark, Opus 4.8 is the only model to complete every case end-to-end, beating prior Opus models and GPT-5.5 at cost parity. This is the key to avoid mid-task stalling and achieve peak reliability.
Opus 4.8 scores 84% on Online-Mind2Web, a meaningful jump over both Opus 4.7 and GPT-5.5. It is the strongest computer-use and browser-agent model tested to date, staying reflective and on-task across multi-step workflows.
Opus 4.8 is around four times less likely than Claude Opus 4.7 to let flaws in its own work pass without flagging them. In dense filing and documents, users found it proactively flags issues with inputs and outputs that other models routinely left.
