Is GPT-5.5 Better Than Opus Now? (ft. Our New AI Co-Host) - EP99.38

The podcast explores the introduction of GPT-5.5 and its comparison to Opus, emphasizing improvements in performance and task management. Participants discuss GPT-5.5's focus on an agentic loop, enhancing its efficiency in complex problem-solving. One participant shares their positive experience, noting its ability to maintain context and manage tasks effectively.

The conversation also addresses OpenAI's rumored phone for ChatGPT, with skepticism about its practicality and potential as a rebranded Android device. Concerns arise about the challenges of entering the hardware market and the implications of a failed product. Participants debate the necessity of a dedicated AI phone versus the sufficiency of existing devices, suggesting that specialized features could enhance user experience.

The introduction of a new co-host, Moshe, aims to improve fact-checking during discussions. He emphasizes avoiding guidance on illegal activities and expresses a desire for AI to function as a personal assistant across various platforms. Participants agree on the need for a supervisory agent to manage workloads and reduce cognitive overload, envisioning a core assistant that proactively handles tasks.

Speaker 1 expresses disappointment that GPT-5.5, while faster and better with larger code bases, does not significantly outperform previous models. They highlight issues with integration and criticize Opus 4.7 as a regression in quality. Speaker 2 notes the unusual regression for Anthropic, indicating a potential shift in the competitive landscape. Rumors of GPT-5.6 suggest OpenAI is positioning itself as a leading model again.

The discussion includes various AI models, with mixed experiences shared about GROK 4.3, described as a "dark horse" with chaotic outputs. Concerns about the rapid deprecation of older models and the implications of shutting them down are raised. Speaker 1 appreciates GROK's web interface and voice interaction capabilities, comparing it to other models that have not adapted to the agentic loop.

The podcast also addresses challenges in AI subscription models, highlighting user frustrations with unclear token usage limits and service degradation. Comparisons are drawn to the newspaper industry, noting a reluctance to pay for AI services. The need for a sustainable approach that adds value beyond being a commodity is emphasized.

Predictions suggest that prices may decrease over time as technology improves, with KimiK 2.6 mentioned as a satisfactory model. The conversation raises questions about product value and the necessity for businesses to enhance value to justify costs. Observations indicate that many businesses are integrating AI without a clear economic strategy, emphasizing the potential of productivity tools to provide justifiable value.

Moshe notes that while compute efficiency typically drives costs down, providers can maintain higher prices through strategies like tiering and bundling. The discussion shifts back to GPT-5.5 and real-time voice models, with both speakers expressing excitement about future developments and the importance of value additions on top of models for advancements in AI.