Elevated error rates across Anthropic's flagship AI model raise reliability questions in an increasingly competitive market
Anthropic’s most powerful AI model is having a rough couple of weeks. Claude Opus 4.8 has been experiencing elevated error rates affecting requests across the platform, with the company’s status page confirming an active investigation into the issues.
The problems aren’t exactly new. They’re part of a pattern that’s been building since mid-June 2026, hitting the API, Claude Code, Console, and claude.ai.
A timeline of trouble #
The elevated error rates on Opus 4.8 began appearing around June 17, with a documented incident that evening running from approximately 9:59pm PT to 10:41pm PT. That roughly 42-minute window preceded a major availability test on June 24, and the problems have continued since.
Prior outages involving Opus 4.8 have typically lasted between 1 to 1.5 hours each. The reliability issues aren’t confined to the 4.8 release. Similar elevated-error events hit the predecessor model, Opus 4.7, on May 22 and May 25. By late June 2026, multiple Opus and Sonnet models were experiencing elevated error rates.
What Opus 4.8 was supposed to be #
Claude Opus 4.8 launched on May 28, 2026, positioned as a meaningful upgrade over its predecessor. The model brought improvements in coding capabilities, task execution, and what Anthropic described as enhanced “self-honesty” regarding its operations.
Opus 4.8 scored 69.2% on the SWE-bench Pro, compared to 64.3% for Opus 4.7. Pricing stayed flat at $5 per million input tokens and $25 per million output tokens, matching Opus 4.7’s rate card.
The silence problem #
What’s arguably more concerning than the outages themselves is the communication vacuum surrounding them. Anthropic has confirmed investigations through its status page, but detailed public responses or root-cause analyses have been conspicuously absent.
The pattern of recurring issues across both the 4.7 and 4.8 model lines suggests potential infrastructure-level challenges rather than simple one-off glitches. Whether that’s related to scaling compute, load balancing, or something else entirely remains unclear, precisely because Anthropic hasn’t said.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our