# What AI costs, said out loud by the people building it.

Markdown copy of https://realalpha.app/costs/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy.

AI cost sheet

104 figures from founders, engineers, investors and hosts, each linked to where it was said. Every row says whether it's their own number, an estimate or something they read.

64 people · 37 shows · checked Nov 4, 2026

## What teams and people spend

31 figures · free to read

| Figure | What it is | Who said it | Where | Kind | Hear it |
| --- | --- | --- | --- | --- | --- |
| $300K a month | Supabase's AI tools, for about 450 people. | **Paul Copplestone** Co-founder & CEO, Supabase | Latent Space Oct 28, 2026 | Own number | [1:12:34](https://realalpha.app/s/supabase-spends-about-300k-a-month-on-ai-tools/#t=4344) |
| $80, then $15 to $20 | Warp's cost per pull request with coding agents, before and after tuning. | **Zach Lloyd** Founder & CEO, Warp | Latent Space Nov 3, 2026 | Own number | [23:57](https://realalpha.app/s/factory-tuning-cut-warps-cost-per-pr-by-about-75/#t=1427) |
| $300M this year | What Salesforce expects to pay Anthropic, mostly for coding. | **Marc Benioff** CEO, Salesforce | All-In May 15, 2026 | Own number | [38:51](https://realalpha.app/s/trusted-enterprise-software-may-hold-up-as-ai-reshapes-saas/#t=2321) |
| Hundreds of thousands a month | Tokens used by some single engineers inside Anthropic. | **Boris Cherny** Head of Claude Code, Anthropic | Lenny's Podcast Feb 19, 2026 | Own number | [27:44](https://realalpha.app/s/give-engineers-tokens-before-optimizing-costs/#t=1654) |
| $3M in two weeks | Overage on one Excel plug-in, at a customer she met. | **May Habib** CEO, Writer | Upstarts Jun 25, 2026 | Own number | [41:50](https://realalpha.app/s/ai-buyers-should-be-held-accountable-for-results/#t=2500) |
| Upwards of $1M a year | What one one-person company spends on Perplexity. | **Aravind Srinivas** Co-founder & CEO, Perplexity | Sourcery Oct 12, 2026 | Own number | [26:52](https://realalpha.app/s/ai-products-compound-through-recursive-usage-not-network/#t=1602) |
| 1x to 2x | Tenex's raw token spend against what it pays its people. | **CJ Hess** Tenex | Latent Space Oct 21, 2026 | Own number | [15:19](https://realalpha.app/s/tenexs-token-spend-is-roughly-one-to-two-times-employee/#t=909) |
| $15 to $20 a run | Tokens for one research run across several models. | **Tobi Lütke** Founder & CEO, Shopify | The Knowledge Project Sep 15, 2026 | Own number | [15:53](https://realalpha.app/s/machines-cant-take-responsibility-so-use-an-ai-council/#t=943) |
| $7,500 to $9,000 a month | Top users at AI-native portfolio companies. The median user is at $200 to $400. | **Sarah Wang** General partner, a16z | a16z Podcast Sep 30, 2026 | Estimate | [21:13](https://realalpha.app/s/ai-adoption-looks-broad-but-is-shallow-with-power-users-far/#t=1263) |
| 0.1x, 1x, 1.5x | A rule for token spend against a salary. Many sit at 0.1, 1 is safe, past 1.5 is trouble. | **Shawn Wang (swyx)** Host, Latent Space | Latent Space Oct 28, 2026 | Estimate | [1:12:06](https://realalpha.app/s/supabase-spends-about-300k-a-month-on-ai-tools/#t=4316) |
| $378 by 1 p.m. | Parag Agrawal's token bill that day, as his host remembers it. | **Joubin Mirzadegan** Partner, Kleiner Perkins | Grit Jul 27, 2026 | Secondhand | [9:13](https://realalpha.app/s/agrawal-built-a-personal-agent-to-catch-dropped-priorities/#t=543) |
| 10,000, then 1,650, then 250 | Anthropic customers spending $100K to $1M, $1M to $10M, and $10M or more. | **Chamath Palihapitiya** Co-host, All-In | All-In Oct 30, 2026 | Secondhand | [27:16](https://realalpha.app/s/anthropics-enterprise-revenue-still-has-room-to-scale/#t=1626) |
| $500K in the bank, $400K a month | His early AI lab's cash against its GPU cluster bill. | **Evan Conrad** Co-founder & CEO, SF Compute | Upstarts Oct 29, 2026 | Own number | [2:38](https://realalpha.app/s/gpu-markets-can-make-compute-easier-to-finance/#t=148) |
| 1% and 10% | AI tool spend against headcount spend, at forward-leaning Fortune 500 companies and at AI-native ones. | **David George** General partner, a16z | a16z Podcast Sep 30, 2026 | Estimate | [21:52](https://realalpha.app/s/ai-adoption-looks-broad-but-is-shallow-with-power-users-far/#t=1302) |
| About $8M a month | What Cerebras spent each month for the 18 months before it could make its wafer-scale chip. | **Andrew Feldman** Co-founder & CEO, Cerebras Systems | Uncapped Sep 15, 2026 | Own number | [11:03](https://realalpha.app/s/failure-analysis-kept-cerebras-going-through-18-months-of/#t=653) |
| $100 to a couple hundred | What Parag Agrawal expected to spend on tokens that day, after tuning his agent. | **Parag Agrawal** Founder & CEO, Parallel | Grit Jul 27, 2026 | Own number | [13:10](https://realalpha.app/s/agrawal-built-a-personal-agent-to-catch-dropped-priorities/#t=780) |
| Over $1,000 a month | What some people inside Anthropic spend on credits, mostly for code migrations. | **Boris Cherny** Head of Claude Code, Anthropic | AI & I Oct 29, 2025 | Own number | [25:20](https://realalpha.app/s/use-parallel-agents-to-migrate-code-and-filter-false/#t=1510) |
| Hundreds of thousands to millions | What running an agent-swarm attack like the Hugging Face one costs. | **Matthew Prince** Co-founder & CEO, Cloudflare | TBPN Oct 22, 2026 | Estimate | [13:00](https://realalpha.app/s/ai-helps-attackers-navigate-systems-after-a-breach/#t=770) |
| $5M a year | What an enterprise might pay OpenAI for forward deployed engineers to build its software. | **Jean-Denis Greze** Town | a16z Podcast Oct 19, 2026 | Estimate | [33:19](https://realalpha.app/s/ai-can-make-custom-software-scale-better-than-fdes/#t=1989) |
| $100M+ a year | Spend on closed models that he says isn't uncommon, and that gets big customers looking at owning theirs. | **Misha Laskin** Co-founder & CEO, Reflection AI | No Priors Oct 9, 2026 | Estimate | [29:42](https://realalpha.app/s/enterprises-go-closed-first-then-move-to-open-models-once/#t=1772) |
| $50 a day | A daily token spend he calls fine for someone who does the work and drives value. | **Parag Agrawal** Founder & CEO, Parallel | Grit Jul 27, 2026 | Estimate | [12:42](https://realalpha.app/s/agrawal-built-a-personal-agent-to-catch-dropped-priorities/#t=752) |
| 0.1x now, maybe 0.5x to 1x | Token spend against a person's salary, and where it could go. | **Joubin Mirzadegan** Partner, Kleiner Perkins | Grit Jul 27, 2026 | Estimate | [13:50](https://realalpha.app/s/agrawal-would-rather-overuse-tokens-than-be-too-cautious/#t=820) |
| $20M to $40M | What an executive may already have spent on an internal company GPT. | **May Habib** CEO, Writer | Upstarts Jun 25, 2026 | Estimate | [9:58](https://realalpha.app/s/internal-politics-determine-how-far-enterprise-ai-projects/#t=588) |
| $20 and $200 a month | What people say the average person and the average employee spend. | **Shawn Wang (swyx)** Host, Latent Space | Latent Space Oct 28, 2026 | Secondhand | [1:11:47](https://realalpha.app/s/supabase-spends-about-300k-a-month-on-ai-tools/#t=4297) |
| $3K to $5K a month, up to hundreds of thousands | OpenAI's 50th percentile and top percentile, as he's heard it. | **Shawn Wang (swyx)** Host, Latent Space | Latent Space Oct 28, 2026 | Secondhand | [1:11:54](https://realalpha.app/s/supabase-spends-about-300k-a-month-on-ai-tools/#t=4304) |
| 0.1% of expenses | Anthropic spend as a share of expenses at the Fortune 500 companies using it. | **Gavin Baker** Managing partner & CIO, Atreides | TBPN Oct 27, 2026 | Secondhand | [1:03](https://realalpha.app/s/anthropics-s-1-shows-a-broader-customer-base-than-expected/#t=53) |
| Under $2,000 | The compute for an agent to beat Minecraft on its own in a little over a day, as the host recalls it. | **Alex Heath** Host, Sources | Sources Oct 21, 2026 | Secondhand | [8:40](https://realalpha.app/s/pichai-expects-a-productivity-leap-from-ai-agents-in-2027/#t=510) |
| $15M of compute | What OpenAI put into solving the Navier-Stokes problem, which carries a $1 million prize. | **Scott Aaronson** Quantum computing researcher | The Peterman Pod Oct 12, 2026 | Secondhand | [16:22](https://realalpha.app/s/openais-15m-navier-stokes-proof-and-how-credit-will-work/#t=972) |
| 8x | Median AI vendor spend at the top 1% of companies against the top 10%, per YipitData. | **Sarah Wang** General partner, a16z | a16z Podcast Sep 30, 2026 | Secondhand | [20:50](https://realalpha.app/s/ai-adoption-looks-broad-but-is-shallow-with-power-users-far/#t=1240) |
| A year's budget in 3 months | How fast Uber burned through its token budget. | **Ed Zitron** Founder & CEO, EZPR | The Diary Of A CEO Aug 27, 2026 | Secondhand | [13:50](https://realalpha.app/s/ai-subscriptions-can-hide-the-true-cost-of-usage/#t=820) |
| $500M in April | What one customer reportedly spent on Claude, per Axios. | **Ed Zitron** Founder & CEO, EZPR | Newcomer Pod Jun 8, 2026 | Secondhand | [29:31](https://realalpha.app/s/anthropic-has-not-defined-its-reported-run-rate/#t=1761) |

## What models cost

23 figures · sign in to see the numbers

| Figure | What it is | Who said it | Where | Kind | Hear it |
| --- | --- | --- | --- | --- | --- |
|  | GPT-5.6 Sol and GLM-5.3 Flash on a benchmark Warp ran, at 95% and 93% correct. | **Zach Lloyd** Founder & CEO, Warp | Latent Space Nov 3, 2026 | Own number | 17:34 |
|  | GPT-6 Luna against GPT-5.6 Sol, while scoring higher on OpenAI's clinician benchmark. | **Karan Singhal** Head of Health AI, OpenAI | a16z Podcast Oct 20, 2026 | Own number | 44:38 |
|  | Fixing a security issue on a top frontier model against CrowdStrike's own, by the seller's count. | **George Kurtz** President & CEO, CrowdStrike | No Priors Oct 22, 2026 | Own number | 21:02 |
|  | What OpenAI did to GPT-5.5's price, and how much customers still save per token because each one does more. | **Sarah Friar** CFO, OpenAI | All-In Jun 2, 2026 | Own number | 18:25 |
|  | Kimi against Sol for security review on Vercel's code, though Kimi finds a bit less. | **Guillermo Rauch** Founder & CEO, Vercel | Knuckle Up with Nakul Aug 25, 2026 | Own number | 50:13 |
|  | How far OpenAI's costs fell from GPT-4 to GPT-5.4, over about two years. | **Sarah Friar** CFO, OpenAI | All-In Jun 2, 2026 | Own number | 18:10 |
|  | A prompt cache hit against fresh tokens, which a cheaper model has to beat to be worth switching to. | **Flo Crivello** Founder & CEO, Lindy | The Cognitive Revolution Aug 10, 2026 | Own number | 1:25:29 |
|  | MuseSpark against Opus for an agent with skill files, in his own use. | **Garry Tan** President & CEO, Y Combinator | Lightcone Jul 29, 2026 | Own number | 16:36 |
|  | What the Navier-Stokes proof could cost to redo, a year after it was done. | **Sam Altman** CEO, OpenAI | Cheeky Pint Oct 29, 2026 | Estimate | 17:46 |
|  | What copying a simple ServiceNow app with a language model would cost, by a customer's count he passes on. | **Bill McDermott** Chairman & CEO, ServiceNow | No Priors Apr 17, 2026 | Secondhand | 23:43 |
|  | Selling a GPU's time to end users against renting it out. | **Evan Conrad** Co-founder & CEO, SF Compute | Upstarts Oct 29, 2026 | Estimate | 38:18 |
|  | What State Street's accounting rebuild was priced at, and what it might cost now. | **Dave Blundin** Founder & GP, Link Ventures | Moonshots Oct 23, 2026 | Estimate | 1:01:11 |
|  | A million tokens on a recent frontier model against Kimi K3 on your own GPUs after a week of engineering. | **Arvind Krishna** Chairman & CEO, IBM | All-In Oct 21, 2026 | Estimate | 12:53 |
|  | How fast the cost of robot tasks is falling. | **Emad Mostaque** Founder, Intelligent Internet | Moonshots Oct 17, 2026 | Estimate | 18:36 |
|  | A million output tokens on open-weight models against closed frontier ones. | **David Friedberg** Co-host, All-In | All-In Sep 28, 2026 | Estimate | 39:26 |
|  | How fast the cost per unit of intelligence falls, at a given level of capability. | **Sam Altman** CEO, OpenAI | BG2 Oct 31, 2025 | Estimate | 20:00 |
|  | A customer service call with a person against one handled by an AI agent. | **Bret Taylor** CEO, Sierra | The Logan Bartlett Show Sep 12, 2025 | Estimate | 12:05 |
|  | What reading a paperback works out to, against talking to a language model. | **Noam Shazeer** Google | Dwarkesh Podcast Feb 12, 2025 | Estimate | 53:52 |
|  | A language model against hiring a software engineer or talking to a doctor or lawyer. | **Noam Shazeer** Google | Dwarkesh Podcast Feb 12, 2025 | Estimate | 54:23 |
|  | What Anthropic reportedly cut its prices by in October, which he says he couldn't confirm. | **Steve Eisman** Investor | Prof G Markets Oct 28, 2026 | Secondhand | 4:53 |
|  | Kimi K3.1 Pro's price per million tokens. | **Peter Diamandis** Host, Moonshots | Moonshots Oct 28, 2026 | Secondhand | 55:39 |
|  | A frontier model's cost to finish one robot subtask, from July to September. | **Peter Diamandis** Host, Moonshots | Moonshots Oct 17, 2026 | Secondhand | 17:52 |
|  | What GPT-6 Astra cost per research-level physics problem. | **Peter Diamandis** Host, Moonshots | Moonshots Oct 17, 2026 | Secondhand | 30:12 |

## Gross margins

11 figures · sign in to see the numbers

| Figure | What it is | Who said it | Where | Kind | Hear it |
| --- | --- | --- | --- | --- | --- |
|  | Anthropic's gross margin on inference, about level with Salesforce's. | **Dylan Patel** Founder & CEO, SemiAnalysis | Big Technology Podcast Oct 9, 2026 | Estimate | 29:45 |
|  | Anthropic's revenue per dollar of compute, from Q4 2025 through Q3 2026. | **Steve Eisman** Investor | Prof G Markets Oct 28, 2026 | Secondhand | 3:48 |
|  | Where Modal's gross margin could settle over the long term. | **Erik Bernhardsson** Co-founder & CEO, Modal | 20VC Nov 2, 2026 | Own number | 16:28 |
|  | Gross margins at traditional inference providers, as the host describes them. | **Harry Stebbings** Host, 20VC | 20VC Nov 2, 2026 | Estimate | 15:34 |
|  | Gross margin at each layer of the AI stack, from applications down to Nvidia. | **Erik Bernhardsson** Co-founder & CEO, Modal | 20VC Nov 2, 2026 | Estimate | 16:12 |
|  | What a Perplexity subscriber pays against what they can run up in tokens. | **Ed Zitron** Founder & CEO, EZPR | The Weekly Show with Jon Stewart Oct 28, 2026 | Estimate | 29:43 |
|  | Fireworks' gross margin now, and where it's expected to reach soon. | **Jason Lemkin** Investor | 20VC Oct 22, 2026 | Estimate | 34:38 |
|  | NVIDIA's gross margin, which he didn't expect to last this long. | **Danny Moses** Host, On The Tape | The Real Eisman Playbook Oct 19, 2026 | Estimate | 11:22 |
|  | A guess at the gross margin on AI labs' inference revenue. | **Gavin Baker** Managing partner & CIO, Atreides | BG2 Jun 11, 2026 | Estimate | 1:05:15 |
|  | The gross margins startups earn on top of model vendors. | **Jared Sleeper** Partner, Avenir | Odd Lots Feb 19, 2026 | Estimate | 29:22 |
|  | Anthropic's cash burn, which he called an upside surprise in its S-1. | **Gavin Baker** Managing partner & CIO, Atreides | TBPN Oct 27, 2026 | Secondhand | 0:48 |

## What buyers pay

15 figures · sign in to see the numbers

| Figure | What it is | Who said it | Where | Kind | Hear it |
| --- | --- | --- | --- | --- | --- |
|  | What one person can burn on ChatGPT's $200 plan, per SemiAnalysis. | **Ed Zitron** Founder & CEO, EZPR | The Diary Of A CEO Aug 27, 2026 | Secondhand | 13:01 |
|  | What OpenAI's investor model assumed developers would pay for agents, a year before. | **Sarah Friar** CFO, OpenAI | All-In Jun 2, 2026 | Own number | 22:03 |
|  | Perplexity's Enterprise Pro and Enterprise Max, with usage billed after the credits run out. | **Aravind Srinivas** Co-founder & CEO, Perplexity | All-In Mar 23, 2026 | Own number | 45:32 |
|  | Lovable's default paid plan. | **Anton Osika** Co-founder & CEO, Lovable | Silicon Valley Girl Sep 22, 2026 | Own number | 23:19 |
|  | What Glean's own studies say it saves, so read it as a seller's claim. | **Arvind Jain** Founder & CEO, Glean | Knuckle Up with Nakul Sep 9, 2026 | Own number | 17:53 |
|  | What Writer says its workflows save on each run, so read it as a seller's claim. | **May Habib** CEO, Writer | Upstarts Jun 25, 2026 | Own number | 12:21 |
|  | What each new Menugen user gets on sign-up, which he says makes image generation a big cost. | **Andrej Karpathy** Anthropic; founder, Eureka Labs | Lightcone Jun 19, 2025 | Own number | 32:03 |
|  | What a monthly check of restaurant wholesaler prices costs to run on Town. | **Jean-Denis Greze** Town | a16z Podcast Oct 19, 2026 | Own number | 26:02 |
|  | What some GitHub Copilot users paid, against the tokens they burned. | **Ed Zitron** Founder & CEO, EZPR | Newcomer Pod Jun 8, 2026 | Estimate | 17:32 |
|  | The flat price he says consumers want, rather than paying by usage. | **David Sacks** Co-founder & partner, Craft Ventures | All-In Apr 17, 2026 | Estimate | 24:25 |
|  | What he expects medical superintelligence to cost its users. | **Mustafa Suleyman** CEO, Microsoft AI | Silicon Valley Girl Nov 14, 2025 | Estimate | 17:12 |
|  | A guess at what Meta's Muse may charge on purchases, on top of Shopify's rates. | **Sean Frank** CEO, Ridge | TBPN Oct 5, 2026 | Estimate | 1:10:37 |
|  | Anthropic's discounts, from its filing amendment. | **John Foley** Head of the Lex column, Financial Times | Prof G Markets Nov 3, 2026 | Secondhand | 4:29 |
|  | Cognition's first Devin plan, as the host recalls it. | **David Senra** Host, Founders | Founders Jun 28, 2026 | Secondhand | 37:38 |
|  | What NetSuite quoted a customer to change its ERP, against what an employee spent building it on Replit. | **Amjad Masad** Founder & CEO, Replit | The Joe Rogan Experience Jul 2, 2025 | Secondhand | 1:42:05 |

## Compute and big bills

24 figures · sign in to see the numbers

| Figure | What it is | Who said it | Where | Kind | Hear it |
| --- | --- | --- | --- | --- | --- |
|  | What one gigawatt of AI data center costs to build. | **Bill Gates** Chair, Gates Foundation | Decoder Oct 22, 2026 | Estimate | 31:54 |
|  | Cash asked on three-to-five-year GPU cluster contracts. | **Evan Conrad** Co-founder & CEO, SF Compute | MTS Oct 28, 2026 | Own number | 3:20 |
|  | Amazon's expected capex for 2026. | **Andy Jassy** CEO, Amazon | a16z Podcast Oct 23, 2026 | Own number | 13:17 |
|  | The commitments Modal sometimes signs with GPU vendors it doesn't yet know well. | **Erik Bernhardsson** Co-founder & CEO, Modal | Latent Space Oct 22, 2026 | Own number | 23:23 |
|  | What Snowflake's Arctic model cost to train. | **Sridhar Ramaswamy** CEO, Snowflake | The Logan Bartlett Show Nov 1, 2024 | Own number | 13:50 |
|  | What he would rather a gigawatt cost over five years. | **Sam Altman** CEO, OpenAI | Cheeky Pint Oct 29, 2026 | Estimate | 39:06 |
|  | What Anthropic is expected to pay Elon Musk's company for compute, from December. | **Ed Zitron** Founder & CEO, EZPR | The Weekly Show with Jon Stewart Oct 28, 2026 | Estimate | 23:38 |
|  | Anthropic's non-cancellable compute commitments with Amazon, Google, Microsoft and Broadcom. | **Ed Zitron** Founder & CEO, EZPR | The Weekly Show with Jon Stewart Oct 28, 2026 | Estimate | 25:31 |
|  | What Anthropic's compute commitments are expected to cost it within a year or two. | **Ed Zitron** Founder & CEO, EZPR | The Weekly Show with Jon Stewart Oct 28, 2026 | Estimate | 26:19 |
|  | What he says is coming, and why only a handful of labs can pay. | **Harry Stebbings** Host, 20VC | 20VC Oct 26, 2026 | Estimate | 16:00 |
|  | TSMC's capex this year against Intel's. | **Dwarkesh Patel** Host, Dwarkesh Podcast | Dwarkesh Podcast Oct 22, 2026 | Estimate | 58:32 |
|  | What running chat on local hardware costs against a data center, then the gap for long-context agents. | **Aravind Srinivas** Co-founder & CEO, Perplexity | Sourcery Oct 12, 2026 | Estimate | 30:39 |
|  | What OpenAI and Anthropic need to spend on training to keep progressing. | **Ed Zitron** Founder & CEO, EZPR | The Diary Of A CEO Aug 27, 2026 | Estimate | 1:49:40 |
|  | Training a model on human data against training on AI data. | **Flo Crivello** Founder & CEO, Lindy | The Cognitive Revolution Aug 10, 2026 | Estimate | 1:42:38 |
|  | Global spending on building data centers. | **Nick Bostrom** Principal researcher, Macrostrategy Research Initiative | The Joe Rogan Experience Jul 14, 2026 | Estimate | 1:25:32 |
|  | The projected budget for AGI next year, set against the Manhattan Project. | **Stuart Russell** Computer science professor, UC Berkeley | The Diary Of A CEO Dec 4, 2025 | Estimate | 15:28 |
|  | Meta's capex this year, and where it could go over the next few years. | **Ed Elson** Host, Prof G Markets | Prof G Markets Oct 26, 2026 | Secondhand | 57:06 |
|  | What Alpaca, a fine-tuned Llama from a small Stanford team, cost to make. | **Alex Atallah** Co-founder, OpenRouter | Sabrina Halper Show Sep 29, 2026 | Secondhand | 6:20 |
|  | What analysts think OpenAI and Anthropic will spend on Microsoft, Google and Amazon. | **Ed Zitron** Founder & CEO, EZPR | The Diary Of A CEO Aug 27, 2026 | Secondhand | 1:48:40 |
|  | The property and equipment big tech has added, including data centers and GPUs. | **Ed Zitron** Founder & CEO, EZPR | The Diary Of A CEO Aug 27, 2026 | Secondhand | 1:52:17 |
|  | What OpenAI says it will put into infrastructure for its models. | **Alex Kantrowitz** Host, Big Technology | Big Technology Podcast Jan 16, 2026 | Secondhand | 4:00 |
|  | What the Stargate initiative was announced to spend on AI computing. | **Karen Hao** Investigative journalist | The Diary Of A CEO Mar 26, 2026 | Secondhand | 1:49:33 |
|  | Public estimates of what OpenAI spends on research experiments. | **Dwarkesh Patel** Host, Dwarkesh Podcast | Dwarkesh Podcast Nov 25, 2025 | Secondhand | 42:00 |
|  | What reinforcement learning cost against training the base model. | **Dwarkesh Patel** Host, Dwarkesh Podcast | Modern Wisdom Aug 11, 2025 | Secondhand | 1:12:41 |

How this sheet is checked

Every number is matched to the transcript of the moment it came from. Where the transcript left any doubt about who was speaking, we checked the video by sight, so a name can still be wrong. Links start a little ahead of the line, though on some episodes our times run off the video's, so one can land early or late. Spot a mistake? Write to [hello@realalpha.tv](mailto:hello@realalpha.tv) and we'll fix the row.