Greg Brockman breaks down 'Spud' — and token demand is going infinite
Friday, April 24, 2026 · 9 min
Two genuinely worthwhile episodes dropped this week, and together they paint a pretty clear picture of where AI is headed in 2026. First, Greg Brockman made a rare pod appearance on Big Technology Podcast for what they called an 'emergency episode' about GPT-5.5 — internally nicknamed 'Spud.' Brockman co-founded OpenAI and stepped back from day-to-day ops, so any time he resurfaces to talk product strategy, it's worth paying attention. The conversation digs into what Spud actually is competitively — a meaningful signal given how crowded the frontier model race has become with Anthropic, Google, and a resurgent open-source ecosystem all nipping at OpenAI's heels. The 'emergency episode' framing suggests the hosts felt this needed to be addressed *now*, which tracks with the buzz around whether GPT-5.5 is a genuine leap or a holding pattern before something bigger. Over on Invest Like the Best, the episode 'Tokenomics, Claude Mythos, and Infinite Demand' is the more intellectually interesting listen of the two. The guest drops a quietly stunning observation: token consumption inside their own team exploded this year in ways nobody anticipated. The framing they use — that ideas are now cheap *and* plentiful, but execution has also gotten cheap, meaning only the *genuinely good* ideas justify any meaningful spend — is a clean, memorable inversion of the classic 'ideas are cheap, execution is everything' startup mantra. That's a real framework shift worth chewing on. If true, it reshapes how we think about where value accrues in the AI economy. On the research side, the chatter this week is around long-context reasoning limits and LLM fingerprinting — technical problems that matter a lot for anyone building on top of these models but haven't broken into mainstream conversation yet. The Qwen3.6–27B local deployment benchmarks (85 TPS on a single RTX 3090) are the kind of 'wait, really?' number that keeps reminding everyone how fast the open-source tier is climbing. Bottom line: Brockman on Spud is the news, but the Invest Like the Best token demand conversation is the one you'll still be thinking about next week.
Listen
Show notes
Two genuinely worthwhile episodes dropped this week, and together they paint a pretty clear picture of where AI is headed in 2026. First, Greg Brockman made a rare pod appearance on Big Technology Podcast for what they called an 'emergency episode' about GPT-5.5 — internally nicknamed 'Spud.' Brockman co-founded OpenAI and stepped back from day-to-day ops, so any time he resurfaces to talk product strategy, it's worth paying attention. The conversation digs into what Spud actually is competitively — a meaningful signal given how crowded the frontier model race has become with Anthropic, Google, and a resurgent open-source ecosystem all nipping at OpenAI's heels. The 'emergency episode' framing suggests the hosts felt this needed to be addressed *now*, which tracks with the buzz around whether GPT-5.5 is a genuine leap or a holding pattern before something bigger. Over on Invest Like the Best, the episode 'Tokenomics, Claude Mythos, and Infinite Demand' is the more intellectually interesting listen of the two. The guest drops a quietly stunning observation: token consumption inside their own team exploded this year in ways nobody anticipated. The framing they use — that ideas are now cheap *and* plentiful, but execution has also gotten cheap, meaning only the *genuinely good* ideas justify any meaningful spend — is a clean, memorable inversion of the classic 'ideas are cheap, execution is everything' startup mantra. That's a real framework shift worth chewing on. If true, it reshapes how we think about where value accrues in the AI economy. On the research side, the chatter this week is around long-context reasoning limits and LLM fingerprinting — technical problems that matter a lot for anyone building on top of these models but haven't broken into mainstream conversation yet. The Qwen3.6–27B local deployment benchmarks (85 TPS on a single RTX 3090) are the kind of 'wait, really?' number that keeps reminding everyone how fast the open-source tier is climbing. Bottom line: Brockman on Spud is the news, but the Invest Like the Best token demand conversation is the one you'll still be thinking about next week.
In this episode
- OpenAI President Greg Brockman on GPT-5.5 “Spud,” AI ... - YouTube — YouTube
OpenAI president and co-founder Greg Brockman joins us to discuss OpenAI's newest model Spud aka GPT 5.5 and where it leaves OpenAI competitively. That's coming up right after this. Welcome to Big Technology Podcast. Today we have an emergency episode with OpenAI president and co-founder Greg Brockman all about GPT 5.5, the famous Spud model, looking at what it does and what it means for OpenAI.…
- Tokenomics, Claude Mythos, and Infinite Demand - YouTube — Invest Like The Best
What used to matter a lot was execution was very very [ __ ] difficult and ideas were cheap. Now ideas are cheap and plentiful but execution is very easy. So really only the good ideas are the ones that can justify the spend on super cheap implementation. You told me this incredible story about how your own team's use of tokens has changed dramatically this year. Yeah. Retell that story and what…
Notable reactions
- Hey, That's My Model! Introducing Chain & Hash, An LLM Fingerprinting Technique [Quick Review] — X
“Hey, That's My Model! Introducing Chain & Hash, An LLM Fingerprinting Technique [Quick Review] Quick review # Hey, That's My Model! Introducing Chain & Hash, An LLM Fingerprinting Technique Mark Russinovich Yanan Cai Ahmed Salem ICLR Apr 23, 2026 References 0 Cited…”
- Scaling Attention via Feature Sparsity [Quick Review] — X
“Scaling Attention via Feature Sparsity [Quick Review] Quick review # Scaling Attention via Feature Sparsity Yan Xie Tiansheng Wen and 5 others ICLR Apr 23, 2026 References 0 Cited by 0 View original Abstract Scaling Transformers to ultra-long contexts is…”
- InftyThink: Breaking the Length Limits of Long-Context Reasoning in Large Language Models [Quick Review] — X
“InftyThink: Breaking the Length Limits of Long-Context Reasoning in Large Language Models [Quick Review] Quick review # InftyThink: Breaking the Length Limits of Long-Context Reasoning in Large Language Models Yuchen Yan Yongliang Shen and 5 others ICLR Apr 23,…”
- An Overnight Stack for Qwen3.6–27B: 85 TPS, 125K Context, Vision — on One RTX 3090 — X
“# An Overnight Stack for Qwen3.6–27B: 85 TPS, 125K Context, Vision — on One RTX 3090 Published: 2026-04-23T13:39:52.807000+00:00 ## Summary The team developed an overnight stack for Qwen3.6–27B: 85 TPS, 125K Context, Vision on One RTX 3090. The team spent six hours…”