I’m not impressed. I just dropped one of my Claude subscriptions and I’m transferring it to another ChatGPT one.
That’s a money decision, so it deserves a plain account. Claude has been relegated to effectively double-checking ChatGPT for the last month. I’m not sure 4.8 is any better than 4.7, and both seem a decided step back from 4.6’s early days.
Opus 4.8 is two days old. Anthropic released it on May 28, saying it “builds on Opus 4.7 with improvements across benchmarks, and is a more effective collaborator,” at the same price as before. The announcement’s own summary, as Simon Willison noted, is that “Users will find Opus 4.8 to be a modest but tangible improvement on its predecessor.” Credit for the candour. A modest improvement on a model that felt like a step back still lands short of where things were.
Others felt a step back too. In late April Anthropic published a post-mortem on the degradation users had been reporting, tracing it to three changes in Claude Code’s defaults and instructions rather than the model itself — among them dropping the default reasoning effort from high to medium on March 4, and an April 16 instruction to keep responses short that “backfired, causing a 3% drop in coding quality evaluations.” Anthropic’s line was “We never intentionally degrade our models.” Fair enough. For the person paying, the effect was the same either way.
Meanwhile GPT-5.5 arrived on April 23, billed by OpenAI as its “smartest and most intuitive to use model” yet, and after a month of it doing the primary work, with Claude checking behind it, the subscription has followed the work.
GPT 5.5 is a real beast.
