Anthropic Launches Claude Opus 4.8 With Gains in Coding and Honesty
Anthropic has launched its latest AI model, Claude Opus 4.8, which features significant improvements in coding and honesty. The new model is reported to be a more effective collaborator, with enhanced capabilities in various tasks and a notable reduction in misaligned behavior. Additionally, Opus 4.8 is faster and more cost-effective than previous versions, while new features are being introduced to enhance user experience.
- ▪Claude Opus 4.8 has improvements in agentic coding, multidisciplinary reasoning, and financial analysis.
- ▪Early testers found the model to be more reliable and less likely to make unsupported claims.
- ▪Opus 4.8 scored 69.2% on SWE-Bench Pro, outperforming GPT–5.5 and Gemini 3.1 Pro in several benchmarks.
Opening excerpt (first ~120 words) tap to expand
Anthropic Launches Claude Opus 4.8 With Gains in Coding and HonestyThursday May 28, 2026 11:29 am PDT by Juli CloverAnthropic today announced the launch of its latest AI model, Claude Opus 4.8. Anthropic claims the model is a "more effective collaborator" with improvements in agentic coding, multidisciplinary reasoning, agentic computer use, knowledge work, and agentic financial analysis. Testers have found Opus 4.8 to be "more reliable and sharper in its judgement" when doing agentic tasks, and the model also made gains in honesty. Early testers report that Opus 4.8 is more likely to flag uncertainties about its work and less likely to make unsupported claims.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at MacRumors.
Discussion
0 commentsMore from MacRumors
