Anthropic releases Claude Opus 4.8 with gains in agentic coding, computer use and 'honesty' metrics
Anthropic released Claude Opus 4.8 on May 28, 2026, an incremental upgrade to April's Opus 4.7, holding pricing flat at $5 per million input tokens and $25 per million output tokens (standard) and $10/$50 for fast mode.
VERDICT — CONFIRMED
Anthropic released Claude Opus 4.8 on May 28, 2026, an incremental upgrade to April's Opus 4.7, holding pricing flat at $5 per million input tokens and $25 per million output tokens (standard) and $10/$50 for fast mode. The company reported across-the-board benchmark gains: agentic coding rising from 64.3% to 69.2%, multidisciplinary reasoning with tools from 54.7% to 57.9%, agentic computer use from 82.8% to 83.4%, agentic financial analysis from 51.5% to 53.9%, and an internal 'knowledge work' score from 1753 to 1890. On named public suites Anthropic cited Terminal-Bench 2.1 at 74.6% and Online-Mind2Web at 84%, and said Opus 4.8 was the first model to break 10% on a legal-agent all-pass standard; the agentic computer-use figure corresponds to OSWorld-Verified at 83.4%.
Anthropic framed the headline improvement as reliability and calibration, stating the model is 'around four times less likely than its predecessor to allow flaws in code it has written to pass unremarked' and is more likely to flag uncertainty rather than make unsupported claims. New product features include 'dynamic workflows' in Claude Code (parallel subagent orchestration for large tasks, shipped as a research preview), user-facing 'effort' controls on claude.ai and Cowork that trade depth against speed and token usage, and Messages API support for mid-task system entries without cache disruption; fast mode runs roughly 2.5x faster and is described as three times cheaper than for prior models. Anthropic also said its Mythos-class cybersecurity model would reach all customers 'in the coming weeks,' pending additional cyber safeguards.
Yahoo Finance noted Opus 4.8 reportedly outperforms OpenAI's GPT-5.5 and Google's Gemini 3.1 Pro on Anthropic's agentic benchmarks, and framed the launch within an accelerating IPO race with OpenAI. Numbers are vendor-reported, though independent trackers including Vellum have since published parallel benchmark results.
