Four posts from the past few days. Anthropic now also marks text from older Claude models, the new message threads in the Claude API fall outside Zero Data Retention, OpenAI reverses a planned reduction in ChatGPT Pro and Images 2.6 makes GIFs in ChatGPT.
Claude: watermark now also on Mythos 5, Opus 4.7 and Sonnet 4.6
According to the Anthropic release notes, text from Claude Mythos 5, Claude Opus 4.7 and Claude Sonnet 4.6 has carried Anthropic's text watermark since 3 November. Previously this applied only to newer models. With Sonnet 4.6, the watermark may be absent if you set `temperature`, `top_p` or `top_k` to a value other than the default. Anthropic does not mention an off switch.
What does this mean for you? If you submit text written with Claude as your own work, expect it to be recognisable as AI-generated. Since this week, the same applies to Grok, as you can read in our blog from yesterday. More about Claude is on the Claude page.
Message threads: faster, but not under ZDR
On 2 November, the beta of message threads arrived in the Claude API. The API then stores the conversation on the server, so you only send the new messages with each request. That shortens the wait for the first response in long agent loops. The documentation states that the feature is not eligible for Zero Data Retention: the conversation, system prompt and tools are held encrypted in memory, are not written to disk, and are gone at the latest an hour after the last use.
What does this mean for you? If you have agreed ZDR with Anthropic because you work with personal data, for example as a lawyer or in healthcare, do not enable message threads unless your processing register and DPIA cover it. You can see how Claude scores in the GDPR check for Claude in software development and the GDPR check for Claude for lawyers.
ChatGPT Pro: reduction scrapped, Work and Codex doubled
OpenAI is scrapping the reduction in Work and Codex credits on ChatGPT Pro that it had announced on 29 September, reports Mixed. Pro 100 and Pro 500 get twice as much Work and Codex usage, and on Pro 200 the lower allowance has been raised to match the higher one. The prices remain $100, $200 and $500 per month.
What does this mean for you? If you use Codex in your work as a developer, you get more room for the same money. Torn between the two big providers? Look at ChatGPT versus Claude.
Images 2.6: making GIFs in ChatGPT
With Images 2.6 you can turn a prompt or photo into an animated GIF in ChatGPT, adjusting the speed and individual frames, The Next Web writes. It applies to all users of ChatGPT and Codex. For developers there are two API models: GPT-Image-2.6 Flare for speed and GPT-Image-2.6 Sunburst for quality. The article does not mention prices.
What does this mean for you? If you create visuals for marketing or graphic design, you can now make short animations yourself in ChatGPT. If you upload photos of people, first check the GDPR check for ChatGPT in marketing to see what happens to your input.
In brief
- Text from Claude Mythos 5, Opus 4.7 and Sonnet 4.6 has carried a watermark since 3 November.
- Message threads in the Claude API are faster, but fall outside ZDR.
- ChatGPT Pro gives more Work and Codex usage instead of less.
- Images 2.6 makes GIFs in ChatGPT and brings two new API models.
- All models and their privacy scores are at models and safety.