Nerd Snipe with Theo and Ben

Theo & Ben are back this week to talk all about how Meta's Muse Wonder is here to impress and it seems like Google still can't make a frontier model! We're breaking it all down, including the random appearance of the Tesla Roadster!

Thank you to Paper & General Translation for Sponsoring!

Sources available on our Substack:

Listen wherever you get your podcasts:

Timestamps:

0:00 Gemini Fanboys Intro

4:31 Gemini 4 Argon Benchmarks vs Reality

16:27 Google's Pricing and Falling Behind

20:30 Grok 4.7 vs Gemini 4

25:01 Google Book and Z Fold Android Pain

36:40 Music Piracy and Computer Use

49:18 Anthropic's Growth and Vibe Coding

59:19 Can AI Replace Human Taste

1:28:17 DeepSeek 1017 and Open Weights

1:34:58 Roadster Flying Car and Wrap Bet

1:51:30 Hunting Tedious Work with Agents

#ai #programming #tech #coding

With OpenAI’s new $500 plan lands alongside reduced subscription value, Theo and Ben explain why cheaper tokens aren’t making DevDay feel like a win, T3 Code's new Orchestrator setup, and more!

Thanks to this episode's sponsor, Posthog:

Listen wherever you get your podcasts:

Sources available on our Substack:

Timestamps:

00:00 OpenAI Dev Day

07:31 T3 Code & Claude Code

20:51 Orchestration V2

30:03 Autoresearch & agent workflows

47:18 AI pricing & API costs

1:19:16 Claude Code & T3 updates

1:35:13 Opus 5.5 vs. GPT-6.1

1:54:24 Dots & proactive AI

2:01:59 More Dev Day features

2:11:30 Grok in Tesla

Theo and Ben break down a packed week of AI releases, from Opus 5.5's insane release to Grok 4.7 falling short of the hype. They then dig into why post-training matters, where Jev, GPT-6 Sol, and Luna fit, and more!

Thank you to General Translation & Paper for Sponsoring!

Sources available on our Substack:

Listen wherever you get your podcasts:

Timestamps:

0:00 Intro

3:18 GPT-6 Sol & Luna

14:58 Grok 4.7

38:47 Jev explained

52:48 Why model routing fails

1:05:06 Opus 5.5

1:26:17 Pricing & usage limits

1:34:23 Building games with Opus

1:41:10 Porting TypeScript to Rust

1:48:57 Agent workflows & testing

Theo & Ben break down Jacob Coxon's resignation after three years in pretraining at OpenAI and Anthropic—and his warning that both labs are racing toward AGI and gambling with our lives—along with Dario's latest blog post about pacing the frontier and how Apple's iPhone Duo could have ruined app development (if AI didn't exist).

Thanks to this episode's sponsor, PostHog:

Listen wherever you get your podcasts:

Sources available on our Substack:

Timestamps:

0:00 Intro

2:53 New iPhones

22:24 Apple and Carriers

32:38 Astra’s Reliability

47:55 Codex Usage and Caching

1:19:44 Anthropic Resignation

1:38:38 Pacing the Frontier

1:40:02 Embedded AI Evaluators

1:43:10 The Hugging Face Hack

2:02:29 React Native Debate

2:09:18 Fast Inference Tradeoffs

2:12:40 Democratic AI Coordination

2:18:22 China and Global Pacing

Theo & Ben break down their latest thoughts on how Astra and Fable 5.1's releases have changed the way they work, how they interact in T3 Code, and then circle back to the latest with Codex usage limits, Gemini 3.8 Flash, Muse Spark 1.3, and the growing gap between model benchmarks and real coding workflows.

Thanks to this episode's sponsor, General Translation:

Listen wherever you get your podcasts:

Sources available on our Substack:

Timestamps:

00:00 Intro

04:57 Gemini 3.8 Flash

14:48 Muse Spark & small models

22:08 AI subscriptions

34:14 GPT-6 Astra launch & pricing

56:04 Rate limits & autonomous coding

1:10:14 Fable 5.1

1:23:23 AI design & 3D demos

1:40:53 Astra vs. Fable: Which to use?

Theo & Ben break down OpenAI's latest model, Astra, and why it's their new benchmark for AI coding, computer use, and multimodal work. But it's not all good: they explain why its UI, stopping behavior, and agent reliability still create friction in real software workflows. From DEFCON puzzles to coding-agent PRs, we're comparing Astra with Fable and asking what a trustworthy OpenAI model should do next.

Thank you to PostHog for sponsoring today's episode!

Listen wherever you get your podcasts:

Sources available on our Substack:

Timestamps

00:00 Meet Astra

04:03 Best Model Ever, With Catches

06:33 Reasoning and 3D Benchmarks

20:00 Astra Rebuilds Ping.gg

30:04 Instruction-Following Problems

44:47 The Uncommitted Fix Debate

01:09:37 The PR Babysitting Failure

01:22:41 Why Fable Still Wins

01:30:41 Multimodal and Computer Use

01:39:29 Astra vs. Fable Fleet Data

Theo & Ben break down OpenAI's GPT-5.6 Sol price cut, breakdown the "Ox Alpha" stealth model we now know is GLM 5.3 Flash, then take a drink every time they say "Grok" while ranking every current AI model on a tier list! What could go wrong?

Thanks to this episode's sponsor, General Translation:

Listen wherever you get your podcasts:

Sources available on our Substack:

Timestamps

0:00 Intro

3:30 OpenAI vs. Anthropic

14:44 Anthropic’s delayed models

20:53 Kimi K3 and open weights

30:08 The Alpha stealth model

48:26 Model tier list begins

1:09:53 GPT-5.6 Luna

1:13:09 Opus and Sonnet

1:20:25 Gemini models

1:31:15 DeepSeek and local models

1:59:58 Fable vs. Sol

2:28:46 Final rankings

Anthropic started watermarking their model's outputs, Dario posted on X, SpaceXAi is getting really good really fast, Meta is shipping again, and we won DEF CON 2026.

Thank you PostHog, the all in one suite of product tools for sponsoring today's Episode!

Check them out at: nerdsnipe.link/posthog

Listen wherever you get your podcasts:

- Spotify: nerdsnipe.link/spotify

- Apple: nerdsnipe.link/apple

- Elsewhere: nerdsnipe.link/listen

Sources available on our Substack: nerdsnipe.substack.com

Timestamps:

0:00 Intro

2:56 Claude Watermarks

15:04 Meta + Muse

48:48 GLM-5.3

55:15 Qwen + DeepSeek

1:08:48 Grok 4.6 + Bot

1:20:15 Gavin vs Dario

1:43:44 DEFCON

2:15:35 Viewer Q&A

Theo and Ben break down why Opus 5 feels like GPT-5.5 crossed with Fable rather than GPT-5.6 crossed with Fable, what Fable found when it audited an Opus thread line by line, where Opus still clearly wins (3D, animation, and Claude Code limits at half Fable's price with no 50% weekly cap), plus Kimi K3, GLM 5.2, the Hugging Face hack, Grok 4.5, Codex, and T3 Code.

Thanks to this episode's sponsor, General Translation:

Sources available on our Substack:

https://nerdsnipe.substack.com/

This week Grok 4.5, Muse Spark 1.1, GPT-Live, and GPT 5.6 all dropped. ⁠Theo⁠⁠⁠ and ⁠⁠Ben⁠⁠ break down which models you should care about, how OpenAI fumbled the Codex to ChatGPT app transition, the abysmal usage numbers for 5.6, and the latest drama surrounding OpenAI and Sam Altman. Plus, how many satellites would it take to trap humanity on Earth, and how many data centers to heat up the ocean?

Thank you to this episode's sponsors: Composio & WorkOs.

Sources available on Substack: https://nerdsnipe.substack.com/

We've spent six figures in tokens testing OpenAI's 5.6 Sol model to see whether if its better than Fable and what OpenAI have done to make it even better than 5.5. Also, we breakdown why we both moved our agents to Linux boxes, how to actually burn $65k on a single loop, and the Codex vs Claude Code subagent gap that's now bigger than the model gap itself.

Thanks to this episode's sponsors Clerk and General Translation:

Sources Available on our Substack:

Fable was supposed to give us 14 days and instead we got 3 before the export ban. Now it's back at half the rate limits. Also, we break down why AI-generated Claude Code skills fall apart, what actually triggers Fable-to-Opus rerouting, the Mythos export-control timeline, and whether GLM 5.2 can really touch Sonnet 5.

Thank you to PostHog for sponsoring today's episode!

Sources available on our Substack:

A new "government-approved rollout" for GPT-5.6 is coming, and it's got us wondering: are we entering the next dark age of AI model access? Additionally, we break down the launch of T3 Code x Grok CLI, Apple's price hikes, the memory supply crunch driving RAM and SSD costs up, a repo-poisoning and AI PR spam wave hitting open source, the GPT-5.6 non-launch, and what frontier model access looks like in a Fable/Mythos-tier world.

Thank you to Composio for sponsoring this episode!

Sources Available on our Substack:

https://nerdsnipe.substack.com/

The US government banned Anthropic's Mythos and Fable models just after launch so we break down exactly how Project Glass Wing, a panicked AWS engineer, and Dario's failure to communicate with Washington triggered the chaos. Plus: SpaceX's $60B Cursor acquisition, GLM 5.2 beating Google, the Codex trick for running 200 parallel agents, and why your next GPU will ship with a GPS tracker

Thank you to Composio and WorkOS for sponsoring this episode!

Sources Available on our Substack: https://nerdsnipe.substack.com/

RIP Fable 5. We recorded this before it got taken offline, but it's still worth talking about. The model is incredible. We really miss it.

Thank you, Firecrawl, Depot, and Clerk for sponsoring!

Sources

Timestamps

Cloudflare buys Void0, Google buying up compute from xAI, and Claude seems to be getting more anxious so we're here to break down everything this week on another episode of Nerd Snipe!

Thank you to Composio for sponsoring today's episode!

Sources:

01:58 Cloudflare vs Vercel

13:03 Convex angle

19:59 SpaceX compute

35:49 AI self-improvement

44:45 Article reactions

50:28 Claude anxiety

57:42 Guardrails

01:08:21 Cursed image gen

01:19:18 Hermes agents

Opus 4.8 + a ton of new stuff in Claude Code dropped this week, and we actually kinda like it. There's also a new benchmark that's actually good, and we have a lot of thoughts about the future of these AI labs...

Thank you PostHog and Clerk for sponsoring!

SOURCES

00:00 - Shower Thoughts

02:44 - Deep SWE Benchmark

10:45 - Opus vs GPT-5.5

19:57 - Anthropic’s Huge Raise

25:39 - Token Maxing

40:02 - AI Slot Machine

43:49 - Claude Code Friction

50:01 - Opus, Mythos, and Safety

Not only did Google accidentally ban Railway's account, but their new flagship model Gemini 3.5 Flash is absurdly bad. Oh and apparently Theo's building his own cloud...

Thank you Macroscope and GT for sponsoring today's episode!

SOURCES

⁠https://x.com/theo/status/2057359424378097823⁠

⁠https://x.com/JustJake/status/2056881510939283776⁠

⁠https://x.com/KorduGG/status/2059141337895604626⁠

⁠https://x.com/unboringtech/status/2059144145273491610⁠

TIMESTAMPS

00:00 - Gemini fallout

05:21 - Video gen

10:10 - Google Cloud

14:04 - Windsurf/Cursor

25:10 - Manus/Meta

30:00 - China lock-in

35:00 - Cursor/xAI

45:00 - Cloud workflows

50:32 - Lakebed

65:22 - Hermes/security

Peter, the creator of openclaw is apparently going through $1,300,000 worth of tokens every month. Seems like we're not using nearly enough tokens. Oh and the security psychosis is getting worse. (Anthropic is being bad again too)

Thank you to today's sponsors!

- AgentMail, email inboxes for your agents: nerdsnipe.link/agentmail

- Clerk, the best experience for auth, orgs, billing, and more: nerdsnipe.link/clerk

TIMESTAMPS

00:00 - Intro

02:44 - Token Spend

06:47 - Token Future

12:12 - Secure Agents

22:50 - Anthropic Rules

31:35 - Security

39:36 - Token Tax

49:49 - AI Psychosis

59:47 - macOS

01:01:43 - AI Video

Anthropic seems to have finally solved their compute problems (kinda) by buying it from Elon, the security problem is getting so much worse, and apparently Bun's getting re-written in rust?

Thank you to PostHog and Composio for sponsoring today's episode!

- PostHog, the all in one suite of product tools: nerdsnipe.link/posthog

- Composio, connect your agents to everything: nerdsnipe.link/composio

Sources/references:

https://x.com/claudeai/status/2052060691893227611

https://openai.com/index/elon-musk-wanted-an-openai-for-profit/#december-2018-elon-told-us-to-raise-billions-per-year-immediately-or-forget-it

https://x.com/jarredsumner/status/2053391824702898475

https://x.com/jarredsumner/status/2051595933704761618

https://x.com/jarredsumner/status/2053047748191232310

https://ze3tar.github.io/post-zcrx.html

https://www.jefftk.com/p/ai-is-breaking-two-vulnerability-cultures

https://x.com/thdxr/status/2053570581807722968

https://x.com/badlogicgames/status/2052691176373805534 https://x.com/garrytan/status/2052996691586932783

TIMESTAMPS

00:00:00 - Anthropic/xAI00:08:59 - OpenAI Lawsuit00:25:40 - Bun in Rust00:34:34 - Security00:59:28 - Pottery Coding01:11:11 - Local Models

This week Theo nearly lost a million dollars and Ben got AI psychosis (from gstack)...

Thank you to Coderabbit and Clerk for sponsoring today's episode!

- Coderabbit, the ultimate AI code reviewer: ⁠nerdsnipe.link/coderabbit

- Clerk, the auth platform with the best DX: nerdsnipe.link/clerk

Sources/references:

- https://x.com/theo/status/2014863266888233193

-https://x.com/theo/status/2050305813894648289

- https://x.com/theo/status/2050314995561611357

- https://x.com/sama/status/2050671161915371998

- https://x.com/davis7/status/2050718508372431026

- https://x.com/thdxr/status/2050719575033983323

- https://x.com/davis7/status/2050762009592148375

- https://x.com/naval/status/2050560057675522500

- https://x.com/theo/status/1952229335416623592

00:00 - Intro / studio return

00:52 - Azure $1M / Microsoft

11:22 - Cloud platform talk

18:08 - Coding agents / SDKs

29:06 - GPT-5.5 pricing debate

45:14 - OpenClaw workflows

01:12:37 - G Stack / G Brain01:29:00 - Dynamic UI / wrap-up

OpenAI and Microsoft are breaking up, Sam's drunk posting, Anthropic is being stupid again, and we still disagree about GPT-5.5Thanks to this episode's sponsors: - Clerk, the auth platform with the best DX: https://nerdsnipe.link/clerk- Coderabbit, the ultimate AI code reviewer: ⁠https://nerdsnipe.link/coderabbit- PlanetScale, the fastest and most scalable cloud databases: https://nerdsnipe.link/planetscale

00:00 Intro

04:05 Sam drunk tweets1

3:13 Anthropic billing woes

28:27 OpenAI divorce

42:34 GitHub can't stop dying

59:40 GPT-5.5 retrospective

Sources/references: https://x.com/sama/status/2046808114561974567

https://x.com/sama/status/2046808217133670800

https://x.com/sama/status/2048160404376105179

https://x.com/sama/status/2047403771416940715

https://x.com/om_patel5/status/2048204411986469232

https://x.com/MSFTnews/status/2048749108127506936

https://x.com/GergelyOrosz/status/2048834949667537369

https://x.com/theo/status/2047721472521621991

https://x.com/mitchellh/status/2049213597419774026

https://x.com/kdaigle/status/2047803291988590609

https://x.com/ryanflorence/status/2048538797638599109

https://x.com/davis7/status/2048239401059434710

https://x.com/davis7/status/2048077518725366173

https://x.com/badlogicgames/status/2048444292562026713

https://x.com/0xSero/status/2048744545853030690

Anthropic month is finally over. We've got a ton to talk about: GPT-5.5 pre-release impressions, the Vercel hack, Cursor + xAI, Qwen models, Kimi k2.6, and so much more...Thank you to today's Sponsors!

Depot, truly modern CI: nerdsnipe.link/depot

Coderabbit, the ultimate AI code reviewer: nerdsnipe.link/coderabbit

Clerk, the auth platform with the best DX: nerdsnipe.link/clerk

TIMESTAMPS00:00:00 - Intro, vacation chaos, and episode setup

00:01:34 - Vercel hack explained and security fallout

00:05:38 - Kimi K2.6 and the rise of strong open-weight models

00:14:29 - Cursor + xAI/SpaceX partnership and acquisition option

00:37:24 - GPT Image 2 impressions, strengths, and flaws

00:43:53 - Anthropic Cloud Code Pro pricing controversy

00:49:30 - GPT-5.5 first impressions: split opinions

01:02:25 - Critiques of GPT-5.5 for coding, context, and prompting

01:22:02 - GPT-5.5 Pro

Really hope that Anthropic month is almost over. But at least for now we gotta talk about Opus 4.7, Claude Design, and much much more...Thank you to this episode's sponsor Clerk! The best DX in auth by a mile: https://nerdsnipe.link/clerkOur RSS feed: https://nerdsnipe.link/rssFollow on spotify and everywhere else you get your podcasts! https://nerdsnipe.link/podTIMESTAMPS00:00:00 Anthropic Again...00:07:30 T3 Code Banned?00:16:08 How Claude Caching Works00:24:00 Why Claude Seems Dumber00:38:03 The Conspiracies Begin...

Diving into Mythos, Claude reasoning, post-AI Uncle Bob, GStack,

Thank you to today's sponsors!

Coderabbit

The AI code reviewer that deeply understands your codebase: ⁠https://nerdsnipe.link/coderabbit⁠

Clerk

The auth platform we always either use, or regret not using: ⁠https://nerdsnipe.link/clerk

00:00:00 - Welcome to the TBPNN Show

00:02:50 - Anthropic's Mythos model

00:11:14 - Anthropic nerfed Claude's reasoning effort

00:14:20 - We should be scared of mythos

00:24:20 - Post-AI Uncle Bob

00:29:34 - Static Typing vs Dynamic Typing and Linting

00:36:23 - Claude Code Skills

00:40:49 - GStack is actually good

00:46:53 - Are we boiling in the ocean?

00:51:07 - Everything should be a md file

00:55:23 - Peace Out Nerds

Talking about everything that went wrong with Anthropic last week (source code leak, rate limits, banning openclaw, and so much more) + our new favorite coding agent Pi.

Thank you to this episode's sponsors!

Coderabbit

The AI code reviewer that deeply understands your codebase: https://nerdsnipe.link/coderabbit

Clerk

The auth platform we always either use, or regret not using: https://nerdsnipe.link/clerk