Opus 5.5 Outperforms GPT-6 Soul in Key AI Tasks

Testing included script generation for a video comparison, with the host setting up both models side-by-side to perform real-time research through built-in browsers.

Paul J. Lipsky conducted a series of tests, including script generation for a video comparison, to evaluate the new models.

Both models used their built-in browsers to perform real-time research, with the host evaluating the quality of the generated scripts and writing style.

Lipsky stated he would run all tests at a high effort level, with GPT-6 Soul on the left and Opus 5.5 on the right.

Under two minutes were required for GPT-6 Soul to complete its research and writing, yet the host found the script from Opus 5.5 clearer and more cohesive.

Opus 5.5 finished slightly later, but its web navigation was notably faster than previous versions of the model.

Lipsky found the script produced by Opus 5.5 to be clearer and more cohesive, emphasizing its human-like tone, making it the preferred model for writing tasks.

Both AI models were tested under a $100 monthly subscription plan, indicating their accessibility for regular use.

The script generation task consumed less than 1% of the weekly usage limit for both services, demonstrating the cost-effectiveness of these models for daily AI work.

Focusing on the Gemini notebook, the host compared GPT-6 Soul and Opus 5.5 against the industry benchmark Fable 5.1 to determine if the models could replicate the motion graphics quality frequently used in his video productions.

Fable 5.1 served as the established benchmark for top-tier motion graphic performance, setting a high standard for the new AI models, with the test focusing on content related to the new Gemini Notebook.

GPT-6 Astra presented an illogical full-screen layout with excessive background space, leading to a subpar visual presentation.

The model failed to track precise click locations for zoom-ins and struggled to properly outline specific sections of the video, resulting in awkward and disjointed zoom transitions.

By identifying and matching the expected video outline, Opus 5.5 maintained an editing style that the host remarked was impressive, specifically noting that he was blown away by its performance.

The model successfully synchronized zoom actions with specific click locations and provided accurate highlighting for key video sections, resulting in an overall natural and consistent editing style.

Both Opus 5.5 and GPT-6 Soul consumed approximately 1% of their usage capacity for the given task, indicating similar resource efficiency.

However, Opus 5.5 completed the task in 7 minutes and 15 seconds, while GPT-6 Soul took 15 minutes and 30 seconds, a surprising outcome given GPT-6 Soul's previous faster benchmarks.

Opus 5.5 is now considered the best available model, outperforming Claude 5.1 in a balance of speed and price, and it has successfully corrected all issues present in its predecessor, Opus 5.

Ultra subscribers now have the ability to query uploaded documents directly through a live chat function, a feature that performed with notable speed during testing.

Users can also reference specific notebooks as sources within Google Docs by using the '@' symbol, enhancing document integration.

The live chat feature was highlighted for its speed and interruptibility during testing, improving user experience.

This new functionality is designed to make reports more dynamic and engaging by integrating a variety of generated assets.

Google introduced a new laptop pre-loaded with Gemini features, including pointer-activated queries and native dictation, showcasing an emphasis on Gemini intelligence.

Gemini's integration capabilities have expanded to 13 new applications, and Google also released Gemini 3.8 text-to-speech, further broadening its reach and functionality.

These developer-focused models are expected to eventually integrate into consumer applications, bringing realistic voice synthesis to a broader audience.

SpaceX AI announced Grok 4.7 as a notable improvement over version 4.6, maintaining the same price and speed, but early user impressions are divided.

The host noted that the update may not significantly impact standard consumer users, suggesting limited practical benefits for the general public.

Grok bot now allows users to route network traffic through their local desktop computer, bypassing blocks associated with cloud virtual machines.

New voice memo and voice chat features enable bots to send audio updates and simulate phone conversations, enhancing interactive capabilities.

Each bot can be assigned a unique voice and speed setting, offering personalized interaction experiences.

The ability to control traffic routing provides greater flexibility and circumvents restrictions often encountered with cloud-based AI services.

New features include dedicated email addresses for communication, live video calls with animated avatars, and Mac app support for remote or automated task control.

Meta clarified that Muse will largely remain free with high usage limits, with monetization primarily coming from transaction fees.

New shopping connectors have been added, including Walmart, Best Buy, Sephora, Wayfair, Shop Pay, and PayPal, while Amazon currently blocks Meta Muse from accessing its site, which is described as a strategic error.

New Ray-Ban Meta audio glasses will offer voice interaction without a camera, and the 'Muse charm,' a compact, fingerprint-secured wearable device, will also facilitate interaction with the AI agent.

The ChatGPT desktop app now supports all Chrome extensions, including password managers, enhancing its utility.

Users can now connect multiple accounts for plugins like Gmail, and ChatGPT voice integrates with email, calendar, and Slack plugins, expanding its communication capabilities. Additionally, ChatGPT can now track and monitor user credit scores.