Anthropic just dropped Claude Opus 4.8, its newest flagship model, and it’s a big leap forward in software development, agentic tasks, complex reasoning, and knowledge work — all while keeping the same price as its predecessor. Over on claude.ai, there’s now a nifty “Effort Control” slider that lets you dial in how deep you want the model to explore on a given task.
Benchmarks tell the story. On SWE-Bench Pro, a standard coding benchmark, Opus 4.8 hits 69.2% — up from 64.3% on the previous version and well ahead of GPT-5.5 (58.6%) and Gemini 3.1 Pro (54.2%). On the multidisciplinary frontier reasoning test Humanity's Last Exam, it scores 57.9% with tool use and 49.8% without, both top of the pack. For web agents, it nails 83.4% on OSWorld-Verified and sets a new industry record of 84.0% on the browser agent test Online-Mind2Web.
Anthropic also introduced a “fast mode” that cranks inference speed up to 2.5x, and API pricing drops to one-third of the previous generation. Early testers report that Opus 4.8 proactively flags potential flaws in input or output data, shows impressive self-correction in long sessions, and generally cleans up the signal-to-noise ratio in development environments.