ChatGPT 5.5: Breaking Down Features

The episode opens with a rapid-fire summary of the week's biggest AI stories before diving deeper into each one. The host begins with SpaceX, which reportedly told investors it is considering manufacturing its own GPUs as part of a 'substantial capital expenditure.' The host interprets this as a signal that the compute bottleneck has become severe enough that even a rocket company is contemplating silicon manufacturing, joining Meta, Google, Amazon, and Apple in pursuing custom chip strategies to reduce dependence on Nvidia.

Next, the host covers Microsoft's integration of Anthropic's Claude Mythos preview into their secure coding framework for threat detection, vulnerability scanning, and incident response. Simultaneously, OpenAI has been briefing federal agencies, state governments, and Five Eyes intelligence partners on a model called GPT 5.5 Cyber. The host is openly skeptical that this briefing effort is substantively different from a PR play designed to mirror the credibility boost Anthropic received when the government engaged seriously with Claude Mythos.

Google's cloud event in Las Vegas is covered next, where the company launched the Gemini enterprise agent platform featuring access to over 200 models through their Model Garden, new model releases including Gemini 3.1 Pro and Flash, a music model called Lyria 3, Gemma 4 on the open-source side, and a management layer for coordinating AI agents with human teams. The host views this as Google's direct response to criticism that it has fallen behind in the enterprise agent race, where OpenAI and Anthropic have been dominant.

The labor disruption segment covers Meta cutting approximately 10% of its workforce — around 18,000 employees — while simultaneously increasing its 2026 CapEx guidance to $115–135 billion, up from $72 billion the prior year. Zuckerberg is cited as explicitly attributing the cuts to AI. Microsoft announced its first-ever voluntary retirement buyout program in its 51-year history, targeting employees whose age plus tenure totals 70 or more, potentially affecting around 8,750 people on top of 9,000 layoffs from the prior summer. Snapchat is also mentioned as cutting 16% of its workforce, with its CEO noting AI now generates over 65% of the company's code. The host frames all of this as the first visible wave of AI-driven white-collar restructuring at major tech companies.

The deepest segment is dedicated to GPT 5.5, released just six weeks after GPT 5.4. The host contextualizes the rapid release cadence against competitive pressure: Anthropic recently hit a $30 billion annualized revenue run rate and is valued higher than OpenAI on secondary markets at $1 trillion versus OpenAI's $850 billion. OpenAI, which is approaching an IPO, is described as unable to take any quarters off. GPT 5.5 is positioned as a fully agentic model designed for multi-step tasks with minimal human direction. Benchmarks cited include 88.7% on SWE-bench, 92.4% on MMLU, 82% on terminal bench, and a claimed 60% reduction in hallucinations versus GPT 5.4. Pricing is $5 per million input tokens and $30 per million output tokens for the standard tier — double GPT 5.4 on paper — though OpenAI claims improved token efficiency makes real-world cost roughly flat. The pro tier is $30 in and $180 out. The host personally notes that OpenAI's upsell approach — prompting users to pay more for 'better responses' — feels less appealing than Anthropic's token-based consumption model.