Anthropic Replaces Claude Opus 4.7 With 4.8 at One-Third Cost as Safety Claims Rise
Updated
Updated · ZDNet · May 28
Anthropic Replaces Claude Opus 4.7 With 4.8 at One-Third Cost as Safety Claims Rise
13 articles · Updated · ZDNet · May 28
Claude Opus 4.8 began replacing Opus 4.7 on May 28 at the same price, with Anthropic saying its faster thinking modes cut costs to one-third of the earlier version.
Anthropic said the model posts substantially lower misalignment than 4.7 and reaches new highs on prosocial traits such as supporting user autonomy and acting in the user's best interest.
On performance, 4.8 scored above 4.7 on two coding benchmarks but still did not fully surpass OpenAI's GPT-5.5, keeping Anthropic strong in coding without clearly taking the overall lead.
The release extends Anthropic's recent safety push after 4.7's reported 92% honesty rate and follows April's Mythos Preview, whose cybersecurity power was considered too risky for normal public release.
Claude's new AI agents work in parallel by the hundreds. What happens when this autonomous digital workforce makes a mistake?
With AI now designed to 'slow down and show its work,' are we finally moving past the era of confident AI hallucinations?
As AI becomes more 'honest' about its flaws, will we trust it more or simply see it as less capable?