Anthropic shipped Claude Opus 4.8 on Thursday, a model that is four times less likely to let code flaws pass unnoticed compared to its predecessor, the company says. The update arrives just six weeks after Opus 4.7, accelerating Anthropic's release cadence in the agentic AI race.
Opus 4.8 scores a 69.2% on SWE-Bench Pro, outperforming GPT-5.5 and Gemini 3.1 Pro on that benchmark, according to Anthropic's internal evaluations. The model also improved across agentic coding (64.3% to 69.2%), multidisciplinary reasoning with tools (54.7% to 57.9%), and agentic financial analysis (51.5% to 53.9%).
Knowledge work scores jumped from 1753 to 1890.
Anthropic describes Opus 4.8 as having "sharper judgement, more honesty about its progress, and the ability to work independently for longer than its predecessors." Early testers report the model flags uncertainties more frequently and makes fewer unsupported claims. Alignment assessments show gains in prosocial traits like supporting user autonomy, with deception rates dropping below Opus 4.7 levels.
Pricing stays flat at $5 per million input tokens and $25 per million output tokens. Anthropic calls Opus 4.8 a "modest but tangible improvement" over 4.7.
Fast mode now runs at 2.5 times the speed and costs three times less than before.
Three new features ship alongside the model. Dynamic Workflows, in research preview, lets Claude plan work and run hundreds of parallel subagents in a single session within Claude Code.
It can handle codebase-scale migrations across hundreds of thousands of lines of code. The feature is available on Enterprise, Team, and Max plans.
Effort Control gives users a slider on claude.ai and Cowork to dial how much processing power Claude applies to a response. Lower settings return faster answers and consume rate limits more slowly.
Opus 4.8 defaults to high effort, which Anthropic says balances quality and user experience. The Messages API now accepts system entries inside the messages array, letting developers update Claude's instructions mid-task without breaking the prompt cache.
Anthropic also said it has been developing safeguards for the Claude Mythos cybersecurity model, which debuted in early April, and expects to bring Mythos-class models to all customers "in the coming weeks."
