Anthropic upgraded its most capable model to Claude Opus 4.8 on 28 May 2026, keeping the price unchanged from the previous version while adding new agentic features and what the company describes as measurable gains in honesty. Anthropic characterised the release as "a modest but tangible improvement on its predecessor", Opus 4.7, rather than a leap in raw capability.
An incremental upgrade at the same price
The new model is available through the Claude API as `claude-opus-4-8`. Pricing is unchanged at US$5 per million input tokens and US$25 per million output tokens. TechCrunch reported that Opus 4.8 arrived 41 days after Opus 4.7, continuing a rapid release cadence. Anthropic said the upgrade improves results across coding, agentic and knowledge-work benchmarks, while framing the gains as incremental.
VentureBeat reported that Opus 4.8 scored 88.6% on SWE-bench Verified, up from 87.6% for Opus 4.7, and 74.6% on Terminal-Bench 2.1, up from 66.1%. Anthropic has published a wider set of evaluations in the model's system card.
A focus on flagging uncertainty
Anthropic placed particular emphasis on the model's honesty. It said Opus 4.8 is "more likely to flag uncertainties about its work and less likely to make unsupported claims", and that the model is around four times less likely than Opus 4.7 to let flaws in its own code pass unremarked.
Its alignment team concluded that Opus 4.8 "reaches new highs on our measures of prosocial traits like supporting user autonomy", with rates of misaligned behaviour substantially lower than Opus 4.7. TechCrunch reported that the investment firm Bridgewater Associates found the model's tendency to proactively flag problems with the inputs and outputs of an analysis set it apart from competing systems.
New features alongside the model
Anthropic shipped several features with the release. Dynamic workflows, available in research preview, lets Claude Code plan a task and run hundreds of parallel subagents in a single session before verifying the output. Anthropic said the feature can carry out codebase-scale migrations across hundreds of thousands of lines of code, using an existing test suite as the bar for success. It is limited to the Enterprise, Team and Max plans.
A new effort control, available on all plans, lets users decide how much work the model puts into a response, trading speed against depth. The Messages API now accepts system instructions inside the messages array, which lets developers update a model's instructions mid-task without breaking the prompt cache. Fast mode, which runs at 2.5 times the speed, is now priced at US$10 per million input tokens and US$50 per million output tokens — which Anthropic said is three times cheaper than fast mode for previous models.
What comes next
Anthropic said it is working on cheaper models with similar capabilities, and on a new class of model more capable than Opus. A small number of organisations are using an unreleased system called Claude Mythos Preview for cybersecurity work under a programme named Project Glasswing. Anthropic said Mythos-class models require stronger cyber safeguards before a wider release, which it expects "in the coming weeks". It has not given a firm date.
