AI Tools 15 November 2026: Claude Projects, Opus 5.5, Seedance 2.5 Flash
Three tools in detail: an ongoing chat that works through tasks in parallel, Anthropic's guide to the default model Opus 5.5, and a fast preview for AI videos. Plus news for online shops and marketing.
Key points at a glance
- Claude Projects bundles many tasks in one ongoing chat and works through them in parallel in the cloud; the beta is available on the Pro, Max and Ultra plans.
- In a guide published on 12 November 2026, Anthropic explains how to prompt the default model Opus 5.5 in everyday office work: more briefly and with a clear goal.
- Higgsfield adds a fast preview variant and 4K to the Seedance 2.5 video model.
Tools of the day – explained in detail
Claude Projects: an ongoing chat that works through tasks in parallel
What is it? Claude Projects is a feature of Anthropic's AI assistant Claude (documentation, as of 15 November 2026; announcement of 17 September 2026). A project is an ongoing conversation. You write in it what needs doing, and Claude starts a separate workstream, a 'thread', for each task. The threads mostly run in the cloud, i.e. on the provider's computers, and keep working when your laptop is closed.
What does it mean? Until now, you had to manage several AI sessions yourself and explain the background to each one again. In a project, a coordinator takes this on: it distributes the work, and an overview shows what is finished and where your answer is needed. Rules you set once apply to every new thread. According to a summary by CellCog, an update on 13 November 2026 brings side questions in their own thread (/btw), branching off a thread (/fork), reminders and more proactive status updates. CellCog also mentions read-only links for sharing; the documentation, however, still lists sharing threads as not possible.
What you can do with it:
- Website maintenance: write error messages and change requests into the chat one after another. If the code is on GitHub, a platform for managing program code, each thread creates its own change proposal for review.
- Online shop: run several jobs at the same time, such as research into competitors' shipping terms, drafts for category texts and an FAQ list. The results are saved as files in the project's library.
- Marketing: for a campaign, start one thread each for the newsletter draft and the landing page copy.
- Office: upload manuals or quotation templates and have them checked for inconsistencies.
- Routine: set up a scheduled task, for example a summary of open items every Monday.
Requirements, costs and limitations: You need a Claude Pro, Max or Ultra subscription. The beta is being rolled out gradually; otherwise there is a waiting list. Team and Enterprise are not yet included. Projects runs in the browser, the desktop app and the mobile app, but not in the terminal. There is no extra charge; the documentation does not mention subscription prices or regions. New projects use the Opus model, with high computing effort ('Effort') in the threads. This uses up the allowance more quickly, especially on the Pro plan; both can be lowered in the project settings. A project belongs to one person; there are no organisation-level controls in the beta. Uploaded files are processed in Anthropic's cloud. Check beforehand whether they contain personal or confidential data.
How to get started: Check at claude.ai/code whether 'Projects' appears in the sidebar. Create a project with a one-sentence goal and start with a small task.
Claude Opus 5.5: the default model and Anthropic's guide to it
What is it? Opus 5.5 is an AI language model from Anthropic and, according to the provider, the default model on the Claude Pro, Max, Ultra and Team plans. On 12 November 2026, Anthropic published a guide on how to use it for research, analysis, writing, spreadsheets and presentations. The Decoder summarised it in German on 14 November 2026.
What does it mean? Familiar tricks become unnecessary. Additions such as 'think hard' should be removed, as according to Anthropic the model reasons anyway. Instead of adding instructions one by one, you state in the first message the goal, the finished result and the point at which Claude should ask you. The effort level ('Effort', i.e. how much computing time goes into an answer) is set to 'Medium' by default. You only increase it if the result is not good enough.
What you can do with it:
- Request a finished file rather than an outline, for example a quotation costing as a spreadsheet.
- Have a long manual or a quotation text checked for errors and inconsistencies.
- Ask an open research question, for instance about competitors, and receive the result as a memo or presentation.
- Attach a chart or a screenshot from your shop statistics instead of typing out figures.
- Have a draft landing page created, supply brand guidelines and name specific design features you do not want.
Requirements, costs and limitations: You need a paid Claude plan; the guide does not mention subscription prices. Via the API, the interface for your own programs, Opus 5.5 costs 4 dollars per million input tokens and 20 dollars per million output tokens. Tokens are the word fragments on which billing is based. The smaller Sonnet 5.5 costs 2 and 10 dollars. The performance figures come from Anthropic's internal tests and customer quotes and have not been independently verified. In one of these tests, 16 of 18 research reports passed the check for invented figures or quotes, meaning two did not. You should therefore verify figures and sources.
How to get started: Search your saved prompt templates for lines such as 'Think carefully' and delete them. Write your next task as a single message with the goal and the desired result.
Seedance 2.5 Flash and 4K on Higgsfield: AI video with fast preview
What is it? Higgsfield is an online platform for AI videos. According to its changelog, Seedance 2.5 Flash, a lighter variant of ByteDance's Seedance 2.5 video model, has been available there since 12 November 2026. It delivers fast previews in 480p or 720p with sound. Since 11 November 2026, Seedance 2.5 has also produced videos in 4K, alongside 480p, 720p and 1080p.
What does it mean? An idea can first be tested at low resolution before the final version is rendered. According to the changelog, in 'Draft Mode' a selected draft can be rendered in 4K. Clips are 4 to 30 seconds long.
What you can do with it:
- Create several preview variants of a product clip for your online shop and render only the best one in final quality.
- Produce a short video with sound for social media, for example to announce a promotion.
- Create a mood video for the header area of your website in 1080p.
- Output a clip in 4K for a trade fair screen or shop window display.
- Show the team a preview for approval before costs are incurred for the final version.
Requirements, costs and limitations: You need a Higgsfield account. The changelog does not mention prices, only that the bitrate setting affects level of detail and cost. The only regional restriction it mentions is the USA, where both are currently unavailable; there is no explicit statement about Germany. Before use, clarify the usage rights under the provider's terms and whether the video must be labelled as AI-generated.
How to get started: Describe a scene in two sentences, generate a Flash preview in 720p and only then render at a higher resolution.
In brief
- Higgsfield GPU (changelog, 13 November 2026): a rentable graphics card in the cloud, controlled via chat through Claude or ChatGPT. Billing is in credits per minute; by default, the rental pauses after ten minutes of inactivity.
- ElevenAI (ElevenLabs; TechCrunch, 13 November 2026): app for meeting notes and dictation in more than 90 languages. So far only a preview with sign-up; the report gives no prices.
- Make (release notes, 13 November 2026): The gemini-4-carbon model can be selected in the 'Simple text prompt' module. The module for Google Ads lead forms provides four new fields, including lead source and time of submission.
Marketing, websites and online shops
AI traffic remains small (Leadfeeder, 28 October 2026; t3n, 13 November 2026): In the B2B data provider's network, 498,439 visits came via AI assistants in July 2026 and 26.1 million via organic search. The AI share has been around 1.8 per cent for about a year. Optimising for AI answers is therefore a complement to SEO, not a replacement.
Google Merchant Center (Search Engine Roundtable, 13 November 2026; Google Help): Promotion titles may no longer contain dates; only a short month abbreviation is allowed. Redemption periods belong in the fields provided for them. Check your Black Friday promotions.
Purchase on invoice and instalments (t3n, 14 November 2026): From 20 November 2026, the new EU Consumer Credit Directive applies in Germany. Payment services such as Klarna or PayPal will then check creditworthiness even for small amounts and are adapting their checkout.
Also important
- Microsoft 365 was disrupted for around 18 hours from 12 November 2026, including Copilot (TechCrunch, 13 November 2026). According to Borncity (14 November 2026), phishing emails with the subject line 'Ausfall – [six-digit code]' (German for 'outage') are exploiting the situation.
- Cursor published an update on its security incident on 13 November 2026: attackers found login credentials in copied session logs. Login credentials do not belong in prompts.
- ChatGPT created 96 per cent of 2,050 memory entries unprompted in a study; deleting a chat does not remove them (t3n, 14 November 2026).
Would you like to use one of these tools for your website, shop or office? Get in touch – we set it up and show you how it works.
This overview is compiled daily with AI assistance from publicly available sources. As of 15 November 2026. Prices and availability may change – the linked sources are authoritative. This article does not constitute legal advice.







