I Tested Grok 4.7, TypeSafe’s Jev and Qwen 3.8 — Here’s What Surprised Me

Robbert van Empel published a new AI-news video on September 22, 2026, bringing together hands-on tests of Grok 4.7, TypeSafe’s Jev and Qwen 3.8 with a wider set of updates from the week. The video is not a single-model benchmark or a vendor announcement. It is a practical tour through the tools, outputs and surprises that become visible when new AI products are used in real workflows.

The first part looks at TypeSafe’s Jev ecosystem. The video discusses JEV Signal for filtering content, Unclutter for cleaning up distracting advertising and JEV Ultra Fast for browser automation. That combination matters because AI tools are increasingly used between the browser, research material and the final task. The examples show the difference between a model that can answer a prompt and a toolchain that helps someone move through a workflow.

Qwen 3.8 Omni Flash is tested through multimodal and creative tasks, including web-design experiments and a South Park-inspired LEGO challenge. These demonstrations make the comparison easier to inspect than a leaderboard alone: viewers can see how the model handles visual context, code and a requested style. They remain Robbert’s own tests rather than a standardized independent benchmark, so prompts, run time and human editing affect the result.

Grok 4.7 gets a practical check of its pricing, availability and performance claims. The video asks what users can access without paying, what changes in paid or developer contexts and whether the model’s capabilities are useful beyond launch messaging. Other short updates cover Anthropic’s Claude Cowork direction, Meta Muse and its availability on Mac, a reported block around Amazon access, Qwen Image 2.1, AI-safety research and new robot hardware.

The value of this video is its combination of inspection and context. It shows where differences between models appear in practice: quality, speed, cost, tool use and the amount of correction a person still needs to provide. It should be watched as a first-party review from the creator, not as proof that one system is universally best. The linked sources in the description let viewers verify the underlying announcements and try the products for themselves.