Anthropic and OpenAI just dropped new artificial intelligence (AI) models, and they both come with a welcome change: the chatbots ramble less.
On Sept 22, Anthropic launched Opus 5.5, the latest addition to the Claude 5.5 family. OpenAI also released two new models, GPT-6 Sol and Luna, the successors to GPT-6 Astra. Both companies are touting broad gains in performance. But one of the most noticeable changes may be how the models communicate.
Anthropic says Opus 5.5 responds more naturally, with writing that is clearer, easier to follow, and more concise without sacrificing accuracy. For founders who spend hours brainstorming, writing, or solving problems with Claude, that could make the chatbot feel more like a capable work partner.
Early testers claim the difference is noticeable. “Verbose, hard-to-follow output has been my biggest frustration with frontier models,” said John Ruelas, a software engineer at Ramp, who used Opus 5.5 to write a software design specification document. He said Opus 5.5 “writes like a good colleague” and “follows our writing rules,” resulting in outputs that are easier to comprehend and require less editing.
OpenAI is tackling many of the same frustrations. The company says GPT-6 Sol and Luna, now available in ChatGPT Work and Codex across its paid plans, give slightly shorter answers with more clarity, less jargon, and fewer unnecessary details compared to GPT-5.6. They are also less repetitive and less likely to jump to conclusions.
One comparison shows the difference. OpenAI asked GPT-6 Sol and an older Sol model to update a website using a “bento box design style.” The older Sol model began with: “Absolutely, I can get the bento feel.” GPT-6 Sol skipped the throat-clearing and got straight to the point: “I’ll reshape the pages into a bento grid.”
Better communication is only part of the upgrade. OpenAI and Anthropic say their new models are also faster, cheaper, and better at complex work, including coding.
The models are the first major releases since several top AI leaders backed calls to slow the pace of AI development. Anthropic CEO Dario Amodei urged tech companies and governments to work together to prevent AI from spinning out of control, citing warnings from researchers that advanced AI could wipe out humanity. OpenAI CEO Sam Altman and SpaceXAI CEO Elon Musk supported Amodei’s push for stronger safety standards. Meanwhile, Nvidia CEO Jensen Huang rejected calls for more regulation.
The new releases raise questions about their commitments to pumping the brakes. If AI companies are promising to slow down, what does that actually mean for the pace at which they launch more powerful models?
“The pacing ‘slowdown’ pledge is a signal, an aspiration,” Thomas Randall, a research director at technology research firm Info-Tech Research Group, tells Inc. “We cannot yet treat the pledges as a reliable indicator for slower product releases, and competitive dynamics remain largely unconstrained.”
Nevertheless, OpenAI and Anthropic promise their most recent products include stronger safeguards. OpenAI calls GPT-6 Sol and GPT-5.6 Sol its most aligned models to date, boosting accuracy and obedience to direction. In testing, agents running on Sol were more likely than their GPT-5.6 counterparts to disclose when search tools were broken, avoid deceptive coding behaviour, and follow only authorised instructions.
Anthropic says Opus 5.5 also performed better on safety tests measuring behaviours such as biased reasoning and attempts to escape containment, concerns that have taken on greater urgency following recent cybersecurity incidents. – Inc./Tribune News Service
