- Official Post
Within three weeks, OpenAI and Anthropic have completely reshuffled their models, which means that almost every older comparison of Claude vs. ChatGPT is outdated. OpenAI unveiled GPT-6 Astra on September 3, and Anthropic followed up with Claude Opus 5.5 on September 22. The short answer up front: Claude for text, documents, and code; ChatGPT for anything involving images and voice. The rest depends on the plan.
Which models are actually included in September 2026?
With ChatGPT, that is no longer so easy to answer. In the regular chat, Astra runs as GPT-6 Pro only on the Pro, Business, and Enterprise plans, while Plus users get Astra only in ChatGPT Work and Codex. In everyday use, paying users work with GPT-5.6 Sol, while Free and Go use GPT-5.6 Luna. If you subscribe to Plus, you therefore do not get the best OpenAI model in chat. There is also some instability: the model selection has changed three times within six weeks.
The situation is clearer with Anthropic. Since September 1, Claude Fable 5.1 has been the most expensive flagship model, with a one-million-token context window. According to Anthropic, the new Opus 5.5 performs at Fable's level on most tasks, costs $4 or $20 per million tokens via the API, and outputs text more than 30 percent faster than Opus 5. Sonnet 5.5 and Haiku 5.5 are expected to follow in the coming weeks.
Claude leads at programming, but by less than the advertising suggests
On paper, the gap is significant. On Terminal-Bench 4.0, Opus 5.5 scores 66.4 percent, compared with 57.9 percent for GPT-6 Astra. In the AutomationBench workflow test, however, Astra leads by 41.4 to 40.0 percent. These are manufacturer figures, though. Even Anthropic acknowledges that benchmark gaps at this level say less and less about real-world differences.
The practical advantage lies more in the package. Claude Pro includes projects, Research, Claude Code, and Claude Cowork, so for $20 you already get a fully fledged coding agent. As an example, Anthropic cites a code migration involving 680,000 lines that was completed in under a day. Showcases like these are, of course, carefully selected. The overall direction is still right.
ChatGPT is simply better at images, voice, and everyday tasks
Here, the matter is clear-cut. On September 8, ChatGPT Images 2.5 arrived with a Sketch feature that converts your own drawings into AI images, with half the waiting time. Claude does not generate images at all. ChatGPT also offers more natural voice conversations, a large catalog of custom GPTs, and, through Microsoft Copilot, a direct route into Office.
So if you create thumbnails, ask questions by voice while on the go, or photograph an invoice and have it analyzed, you are better off with ChatGPT. Period.
Who writes better German?
The difference is narrower than many comparisons claim. Both handle German very well for writing, summarizing, and translation; for professional translations, DeepL is often still a level above. With long texts subject to strict style guidelines, Claude has the edge in our experience. Anthropic explicitly emphasizes that Opus 5.5 puts important information first, uses less technical jargon, and follows prescribed writing rules. Anyone who asks it to write a blog article with ten style rules in the prompt will notice this quickly.
What do the subscriptions cost in Germany?
The pricing structure is almost identical when it comes to ChatGPT vs. Claude:
| Plan | ChatGPT | Claude |
|---|---|---|
| Entry-level | Free, Go for $8 | Free |
| Standard | Plus, €23 | Pro, approx. €20 |
| Heavy users | Pro from approx. €100 to €200 | Max from approx. €100 to €200 |
ChatGPT Plus costs €23 per month including VAT, while Claude Pro costs around €20. The only real difference is at the bottom end: OpenAI offers an $8 tier with Go, while Anthropic has nothing between Free and Pro.
Who is each assistant not suitable for?
Claude is not an all-purpose tool. There is no image generation, no affordable intermediate plan, and intensive users run into the five-hour limits. Anthropic has raised those limits for Pro, Max, and Team with Opus 5.5, but they have not disappeared.
ChatGPT, on the other hand, is not for people who want to know exactly which model is answering at any given moment. The separation between Chat, Work, and Codex is confusing, and the flagship model remains unavailable in the Plus chat. OpenAI is still interesting for developers, though: we are building our support assistant for Aivor on the Responses API.
What comes next
October will be exciting. If Sonnet 5.5 arrives at Anthropic and OpenAI brings GPT-6 into the regular chat, the picture will shift again. We will update the article then.
Over to you: What are you currently working with, and what task made you switch? I am especially interested in cases where one of the two completely failed for you.
P.S.
Aivor Chat automatically selects the best available AI model, can be trained with your knowledge, and can even be easily embedded on your own website. Find out more here:
This post has been automatically translated.