AI News: ChatGPT Ultrafast, Grok 4.6, 3 New Open-Source Models, and more!
Matthew Berman showed 8 AI tools in this video on . 3 are verified available in the EU. Each row below jumps to the exact second. Part of week 2026-33.
Tools shown in this video, with timestamps
Preview of a fast mode for GPT-5.6 Soul running on Cerebras chips at about 14-15x speed, at extra cost. In a demo a financial terminal dashboard finished in 1:50 versus 12:20 on regular Soul; the bottleneck shifts to tool calling and CPUs.
Live demoAnthropic is embedding hidden statistical watermarks in Claude text output, via key-based word choice, to comply with the EU AI Act transparency code. The creator is skeptical about the impact on outputs; it applies to future Claude models.
New xAI model, close to GPT-5.6 Soul and Fable 5 but not quite equal. Priced at $2 per million input and $6 per million output tokens, and fast.
A simplified coding-agent style product with no visible code or model choices. Each thread is an agent, it has plugins like Slack, Google Docs and email, and agents can talk to each other. The creator compares it to a more polished OpenClaw and has used it for a week.
Live demoOpen-source model with a big improvement over GLM 5.2 on Terminal Bench (66.9), beating Kimi K3 on one benchmark but behind Fable and Soul. Very cheap and competitive.
Open model scoring 87.9 on Terminal Bench, close to Kimi K3 and Fable 5. Very low prices with off-peak discounts: $0.66 per million input and $1.98 output, and $0.02 per million for cache hits.
Meta's open-weights 30B-parameter model meant for on-device use, fitting a desktop GPU. It scores 51 on Terminal Bench, well below frontier models, and the creator is glad Meta is releasing open models again.
Opt-in feature that records what you do on your computer to suggest automations, with fine-grained controls over which apps are shared. The creator compares it to Microsoft Recall and says he'd probably try it.
