Merlin

Merlin's bet is that the browser tab is where AI work actually happens, and the product is built around that assumption rather than around a chat window. The Ctrl/⌘+M shortcut reads whatever you're looking at — a long article, a lecture PDF, a PPT deck, a YouTube video — and answers in place, which removes the copy-paste-tab-switch loop that eats time on research-heavy days. The site-specific buttons are the practical part: on Gmail it composes and replies using the previous mail as context, on X and LinkedIn it drafts posts from your saved prompt library, and on YouTube it turns a video into a blog, X post or article. For anyone whose job is repurposing content or clearing an inbox, that's the value, not the model roster. The model roster is the marketing hook, and it's genuinely wide for one account: the vendor lists o3, o4-mini, GPT 4.1, Claude 3.7 Sonnet with thinking, DeepSeek R1, Gemini 2.5 Pro and Flash, Mistral Large, Llama 3.1 405B and Grok 3 — roughly $130/mo of separate subscriptions for $19/mo if you take the annual rate, which is how the vendor frames it. The second layer is where it differentiates from thin chat wrappers: Projects builds reusable knowledge bases and custom bots from your own PDFs and guidebooks, Crafts generates flowcharts, mindmaps and 20+ infographic types from a prompt, Live Search puts reasoning models on realtime web sources, and image generation spans 20+ image models including Flux 1.1 Pro. Add XLS/CSV uploads for quick data reads and 25+ language translation, and there's real surface area here. Where it thins out is control and collaboration. Customization options are minimal, file handling is bounded by what Projects supports, and there's no shared workspace with comments or co-editing — a team of five gets per-seat billing, not a collaborative surface. Teams that have standardized on one vendor's ecosystem will also find Merlin's features trail that vendor's first-party releases, because Merlin is a layer over third-party models and tracks their behaviour. If you need API access, fine-grained parameters or offline use, this is the wrong shape of product. Weigh it against going direct. ChatGPT and Claude have their own browser and desktop surfaces and ship their newest models to their own apps first; Copilot comes bundled with Windows and Edge. Merlin's counter is aggregation plus a page-context workflow you don't have to build yourself, along with availability in 200+ countries including regions where other LLMs are restricted — a real differentiator the vendor calls out explicitly. Treat it as a convenience layer with honest limits, not as a replacement for a primary AI subscription if depth matters to you.