Model‑update velocity rises
A widely viewed Apr. 13 video bundled reports of leaked Claude Opus 4.7, an Anthropic full‑stack app push, a new GPT model, and M2.7, signalling that model releases and product updates are arriving faster than governance coverage. The clip’s title and timing suggest competition is now about model capability plus tooling and packaging, not models alone. (youtube.com)
Artificial intelligence companies spent the week shipping product changes faster than regulators or standards groups published new rules. A widely watched April 12 video stitched together a Claude leak, Anthropic app features, an OpenAI model release and MiniMax’s M2.7 launch into one news cycle. (youtube.com) The strongest confirmed release in that bundle came from OpenAI on March 5, when it launched GPT-5.4 in ChatGPT, the application programming interface and Codex. OpenAI said the model adds native computer use, supports up to 1 million tokens of context and improves tool search across software connectors. (openai.com) OpenAI also split its paid tier again in April 2026. Its help center says ChatGPT Pro now includes a $100 monthly plan with 5 times the usage of Plus, while the existing $200 Pro plan remains the highest-usage option at 20 times Plus. (help.openai.com) Anthropic’s public release notes show the same shift from chatbot to work software. On March 25, Claude mobile added interactive apps; on March 23, Pro and Max users got computer use in Claude Code and Cowork; on April 9, Anthropic said Claude Cowork reached general availability on macOS and Windows. (support.claude.com) Anthropic’s developer changelog moved just as quickly. Claude Code version 2.1.101, published April 10, added a `/team-onboarding` command and changed remote-session setup so some cloud environments are created automatically instead of requiring web setup first. (code.claude.com) The leak at the center of the video is not yet an official Anthropic announcement. Anthropic’s newsroom still lists Claude Opus 4.6, released February 5, 2026, as its latest Opus update, and no public Anthropic page reviewed here confirms an Opus 4.7 launch. (anthropic.com ) MiniMax, by contrast, has published details directly. Its March 18 post described M2.7 as a model that helped build parts of its own training workflow, and its GitHub repository says the model can use agent teams, dynamic tool search and complex skills for software and office tasks. (minimax.io) (github.com) MiniMax attached benchmark numbers to that claim. The company says M2.7 scored 56.22% on SWE-Pro, 57.0% on Terminal Bench 2 and 1495 Elo on GDPval-AA, while also supporting Word, Excel and PowerPoint editing. (github.com) (minimax.io) OpenAI and Anthropic are making similar packaging moves around images and office work. OpenAI said in December that ChatGPT Images, powered by GPT Image 1.5, makes edits up to four times faster, while Anthropic’s March notes said Claude can create charts, diagrams and other visualizations inline and work across Excel and PowerPoint with shared conversation context. (openai.com) (support.claude.com) That leaves one clear pattern by April 13, 2026: the competition is no longer just about who posts the best benchmark. The public record now shows model launches, pricing changes, coding tools, office integrations and computer-use features arriving within weeks of one another across the same handful of companies. (openai.com) (support.claude.com) (code.claude.com)
