Anthropic released Claude Opus 4.8 on Thursday, an incremental upgrade over Opus 4.7 that the company pitches less on raw benchmark gains and more on judgment. The headline claim from its announcement: the model is roughly four times less likely to let flaws in its own code slip by unflagged.
That figure comes from Anthropic's own evaluations, laid out in the system card, so treat it as self-reported until outside testing weighs in. The angle is honesty. Anthropic says Opus 4.8 flags uncertainty instead of confidently claiming progress it hasn't made, a failure mode that dogs most frontier models.
One early tester, a staff engineer, said the model "catches its own mistakes" and pushes back when a plan looks shaky. Anthropic picked the quotes, so calibrate accordingly.
Two features ship alongside it. Claude Code gets dynamic workflows, now in research preview, letting the agent spin up hundreds of parallel subagents and run codebase-scale migrations from kickoff to merge. Users on claude.ai and Cowork also get an effort control: crank it up for deeper reasoning, down for faster replies. It's on all plans.
Pricing holds at $5 per million input tokens and $25 per million output. Fast mode, which runs at 2.5x speed, now costs three times less than on earlier models. The model is live via the API as claude-opus-4-8.
Bottom Line
Opus 4.8 keeps Opus 4.7 pricing at $5/$25 per million tokens and ships with dynamic workflows in Claude Code, currently in research preview.
Quick Facts
- Released May 28, 2026
- Pricing: $5 per million input tokens, $25 per million output (unchanged from Opus 4.7)
- Fast mode: $10/$50 per million tokens, runs at 2.5x speed
- Around 4x less likely to let code flaws pass unflagged, per Anthropic's own evals
- API model string: claude-opus-4-8
