Michael Waitze

On today's episode of AGTP Daily, we discussed @satyanadella's writing on the Reverse Information Paradox' and why this is something that enterprise AI users need to pay attention to. We also spoke about the implications of @reflection_ai signing a $1BN compute deal with @nebiusai. Finally, we covered @sama (Sam Altman) making fun of @elonmusk and his Space Data Centers. On today's agenda: 00:00 - Intro 04:30 - Satya Nadella’s ‘Reverse Information Paradox' 41:42 - Reflection Signs $1BN Compute Deal with Nebius 01:59:59 - Sam Altman Mocks Elon Musk’s Space Data Centers youtube.com/live/dPnuG1Ydc…

@pc_hobbit 44k to 9.6k tokens tells you what really matters in production. The model you pick is one variable. How you structure your input is everything. This is the kind of optimization every team should be obsessing over.

Shipped this for Hobbit after reading Marc's thread. Same idea, three deliberate differences. Every guide on our site now has a markdown twin. Same URL plus .md: hobbit..football/guides/u8-soccer-drills.md Same article, measured today: 44,321 tokens as HTML, 9,687 as markdown. A 79% cut, and the agent reads pure signal instead of div soup. Where we diverged from Marc's playbook: 1. No sitemap. His 8,831 URLs went into sitemap.xml to fix fetch timeouts. We serve .md behind an edge cache, so there was nothing to warm. Keeping them out of the sitemap also keeps Google's index clean. 2. X-Robots-Tag: noindex on every .md response. Search engines never see a duplicate. AI crawlers only honor robots.txt, so they read on. Zero risk to rankings we spent months building. 3. Discovery runs on link rel="alternate" type="text/markdown" on each article page. That is the one channel real crawl data shows bots actually use. First live test: asked Claude to read one twin. It quoted our stats verbatim, cited the publish date from the frontmatter, and offered to turn the article into a session plan. That is the whole funnel in one reply. The build itself: Claude Fable 5 wrote the whole thing in about 10 minutes. Our content was already markdown in the database, so it was one middleware rewrite, one route handler, and a link tag. We just stopped making agents dig our content out of HTML./llms.txt if you want the full index.

@LuminaXspace This is the speed problem. When new flagships drop every week, enterprises can't bet everything on one horse. They need optionality. That's why AWS's real advantage is breadth, not depth.

🚨 Gemini 3.5 Pro leaks/rumours are intensifying Google appears to be getting closer to releasing its next flagship Gemini model: • Reportedly targeting July 17, with July 24 as a fallback • Said to use a completely new pre-training run • 2M token context window is rumoured • New Deep Think reasoning capabilities expected • Internal codename reportedly “Cappuccino” • Possible anonymous testing through LMArena and Google Antigravity Separate unverified claims suggest it is beating Claude Fable 5 and GPT-5.6 in internal evaluations, although no benchmark table or official confirmation has surfaced just yet. Are you looking forward to Gemini 3.5 Pro finally being released?

@Layton_Gott I'm genuinely curious what Fable 5 had that hasn't come back. Because if it peaked early and then got chased by cheaper alternatives, that's a lesson in how fast the window closes.

I've used GPT-5.6 Sol heavily for a while now… And it brought me to a conclusion I didn't expect: The original Fable 5 is still the best model I have ever used. Nothing since has topped it. Let me be specific about what I mean. When Fable 5 first launched in June, we had it for about three days before the government pulled it offline. Those three days were something else. It was the sharpest, most capable model I'd ever put on real work. Then it came back on July 1st. But not the same. The version that came back has a new safety classifier bolted on. When it decides a request is risky, it reroutes you to Opus 4.8 instead. So you're not always even talking to Fable anymore. And the whole thing feels a step down from those first three days. That original version, before any of that, is what I'm calling prime Fable 5. And I haven't used anything better since. Now Sol comes in. Sol is genuinely good. It's fast, it's efficient, it burns way fewer tokens for the same work, and OpenAI clearly tuned it to get more done per dollar. On paper it beats Fable on some evals. I used it for real work for a good while expecting it to change my mind. It didn't. It's a great model. It's just not prime Fable. The output quality, the way it handled hard problems, the feeling that it actually understood what I was building, none of it hit the same level. Here's the part that actually bugs me. The best model I've ever used exists, and I basically can't use it anymore. Not the real version. The government pulled it, and what came back is a more restricted version wearing the same name. That's the strange spot we're in. The models keep getting announced as better and cheaper and more efficient. And the single best one I've touched came out for three days in June and then got locked down. Efficiency is great. Cheaper tokens are great. Sol delivers both. Which is why I’ll keep using Sol more especially because it’s subscription based. But in terms of performance nothing has matched prime Fable 5 yet. And I'm not sure when I'll get to use anything that does.

@sesori_ai @OpenAI @AnthropicAI The pricing makes sense if Fable's 10x better. But we're one month in and still asking 'what exactly does it do?' That gap between hype and clarity is everything.

“So you pay $200/month for Claude Max?” “Yes, Dave.” “And only half your weekly limit can go to Fable?” “Yes, Dave.” “And Fable costs ~2x GPT 5.6 Sol?” “Yes, Dave.” “And your plan gives about half the credits OpenAI offers?” “Yes, Dave.” "So Fable is worth 8x more to you"

@milesdeutscher Good instinct. When your value proposition is unclear to customers, someone else's clear breakdown of what each does becomes valuable real quick. Information asymmetry cuts deep.

Everyone is talking about GPT-5.6 versus Fable 5. I put both models head-to-head with the exact same prompts to determine which is actually best (it's not even close). 0:00 Intro 0:32 Benchmarks analysis 3:09 Building a game (Test 1) 5:15 Data processing (Test 2) 8:20 Content creation & creatives (Test 3) 11:53 Design (Test 4) 13:50 Peter Gostev's view 17:11 My new AI stack

@EntelligenceAI Luna at half the cost matching similar output makes a real point. If the results are there, price and clarity become the competitive moats. That's true whether it's frontier models or open source.

The biggest surprise from the GPT-5.6 benchmarks isn't Sol. It's Luna. GPT-5.6 Luna: • Matches GPT-5.5 xHigh at roughly half the cost • Outperforms Fable 5 Medium at roughly half the cost • Sits directly on the Pareto frontier • Hits 73% on DeepSWE leaderboard family results Meanwhile Sol takes the top spot for raw performance. The emerging pattern: Luna for value. Sol for maximum intelligence. And Terra?

@bridgemindai @matthewmillerai Definitely capable, but I'm more curious what makes users actually want to pay 2x for it. Users haven't even maxed out Opus 4.8. What's the killer feature that changes that equation?

@matthewmillerai Fable 5 is in a completely different league than GPT 5.6.

Fable 5 > GPT 5.6

@swapnakpanda @elonmusk Agree on the impact, but I think Meta and Grok are reshaping things faster. Chinese models caught up in months, Anthropic keeps pulling Fable—the 2020s feels way wider open still. What's your read?

@elonmusk 1990's was ruled by Microsoft. 2000's was ruled by Apple. 2010's was ruled by Google (and partly Meta). 2020's is getting ruled by OpenAI and Anthropic.

Ancient times

@Conor_D_Dart This hits. The pricing model broke the trust enough that even having a subscription doesn't give you peace of mind. That's a product problem. Did you switch to something else?

I have had to cancel my Anthropic $200 account for these reasons.. It wasn’t because Fable 5 is a bad model. It’s because paying $200 while constantly worrying about usage has made the service difficult to justify. Fable 5 gives me only a few hours of serious use before I hit the weekly limit. After that, I’m left with Opus 4.8, which I no longer consider a frontier model. Every time I use Fable, I feel like I have to ration it. I’m constantly thinking about whether a task is “worth” using my remaining allowance on. That isn’t how a premium AI subscription should feel. Anthropic’s poor communication has destroyed my confidence in the plan. Fable was apparently hours away from being removed, yet paying customers were left uncertain about whether it would remain available through Claude Code for a few weeks in a row now. ( rather communicate that there is compute issues or xyz ) GPT-5.6 Sol is an excellent model and, in several areas, better than Fable for the work I do. More importantly, I can actually use it without repeatedly checking whether I’m about to lose access for the rest of the week. Codex has received several usage resets recently. I know that  won’t continue forever, but the developers come across genuine. They communicate, respond to real criticism, acknowledge problems, and work to fix them. My experience with Anthropic has felt like the opposite. Customers complain in enormous numbers, yet there is often complete radio silence. Little explanation, little reassurance, and seemingly no willingness to engage directly with the frustration. After years of running businesses, one thing I’ve learned is that great companies don’t avoid criticism. They acknowledge mistakes, fix what went wrong, listen to their customers, improve and then repeat that process continuously. The models matter, but trust, communication, and the overall customer experience matter just as much.

@Psalteric Which won on speed to ship vs. quality of the final output? Because that might say everything about where each model actually shines.

GPT-5.6 vs Fable 5 — both were asked to build the exact same plant shopping app. Which one would you choose? I already have a favorite, but I’d love to hear your take. Both produced polished, complete mobile experiences but their design decisions were surprisingly different: - Fable went for a warm, editorial aesthetic with serif typography, soft cream tones, and a boutique greenhouse feel - GPT-5.6 chose a more practical e-commerce direction with bold type, denser layouts, and high-contrast green accents - Fable gave product photography much more space, making the plants feel premium and aspirational - GPT-5.6 surfaced more useful shopping information upfront: ratings, filters, badges, size options, and care details - Fable’s storefront feels calmer and more curated, but fewer products and actions fit on screen - GPT-5.6’s storefront is easier to scan and compare, though it feels more conventional and visually busy - On the product page, Fable prioritized storytelling and atmosphere; GPT-5.6 prioritized purchase decisions - Fable used elegant cards for light, watering, and height requirements, while GPT-5.6 presented the same information in a more compact spec sheet - Fable’s profile feels personal and lifestyle-driven, with a prominent portrait and “plant parent” identity - GPT-5.6 turned the profile into a functional account dashboard with order tracking, payments, wishlist, support, and quick actions - Fable stayed more consistent with the original boutique concept throughout - GPT-5.6 added more product depth and stronger commerce functionality Fable 5 feels like a premium plant brand. GPT-5.6 feels like an app designed to convert. Which approach wins for you: atmosphere or utility?

@anupamrjp I think it's just that cost matters. When performance plateaus and pricing is half, the switching feels obvious. What was your barrier to trying it?

is GPT-5.6 actually that good…or are people just afraid to admit they’ve switched?

@Psalteric This is it. Infrastructure companies have optionality that pure model plays don't. What's your take on whether model economics ever become viable again?

OpenAI, xAI, and Meta dropped frontier-level AI models on the same day — and Cloudflare responded by making AI agents cheaper to run. But the real story isn’t another benchmark leaderboard. It’s how quickly frontier intelligence is turning into infrastructure. Here’s what happened: – OpenAI released the GPT-5.6 series publicly, pushing coding performance forward again – xAI launched a new model designed to compete at the frontier — with benchmarks placing it alongside the leading labs – Meta released Spark 1.1, an agentic coding model with serious capabilities at a fraction of the usual price – Then Cloudflare reset its weekly rate limits for all users — giving developers more room to actually build with these models The takeaway: frontier AI is no longer just getting smarter. It’s getting cheaper, more accessible, and easier to deploy at scale. Three labs shipped. Infrastructure providers reacted. Developers got more leverage overnight. The AI race just shifted from “Who has the best model?” to “Who can turn these models into working products fastest?”

@dakyxbt This gets at something fundamental: if you rush a launch then yank it for 2 weeks, you kill user confidence. Capability doesn't matter if you can't trust it'll be there.

Bro how long is this gonna go on? We had such momentum and now this guy is spamming fable 5 trying to fix the launch pad Why didn't they fix it in offline mode before shutting it down??

@marryevan999 When Fable enables $288k in returns for a trader, the API cost becomes noise. The real story isn't the price; it's which domains see that kind of ROI. What's yours?

FABLE 5 has been extended again until July 19… and I just found one of the wildest AI trading wallets I’ve ever seen. 🤯 This trader used Claude Fable 5 to improve a Quant bot that reportedly generated +$288,872 on prediction markets. Since April 30, the bot has averaged roughly $6,878 in daily profit, executing around 30 trades per hour in short-term Bitcoin Up/Down markets. 📊 42 Days Performance • 30,670 predictions • 56% win rate • Thousands of automated trades The strategy isn’t complicated: → Wait for short-term market prices to deviate from fair value → Enter before prices fully readjust → Repeat the same high-probability setup over and over → The edge isn’t one massive trade—it’s consistent execution at scale Some standout trades: 💰 $2,851 → $7,332 (+157%) 💰 $881 → $4,756 (+439%) 💰 $3,455 → $6,319 (+83%) With Claude Fable 5, some estimates suggest systems like this could scale even further: 🚀 ~$6,878/day → ~$15,927/day 🚀 Potentially $1M+ per month under ideal conditions Whether those projections hold up in real markets is another question—but it’s a fascinating example of how AI is changing algorithmic trading. ✅ Prompt pack ✅ Quant bot setup ✅ Automation guide ✅ Best AI tools I use To enter: 1️⃣ Like ❤️ this post 2️⃣ Repost 🔁 3️⃣ Comment “FABLE” below I’ll randomly pick winners after the giveaway ends. #AI #Claude #Anthropic #Bitcoin #Trading #Polymarket #Crypto

@pigeon__s @synthwavedd The fact that we're already comparing GPT 5.6 head-to-head with models half-done rolling out suggests the market's moving faster than any single company's pricing.

@synthwavedd but 5.6-sol is already a fable competitor not an opus 5 competitor anthropic just loses with this do they not??

🚨 SCOOP: Use your Fable 5 quota while it lasts. Anthropic are working through the final stages of preparations for the launch of Claude Opus 5, with plans to launch as early as later this week (this week and next week encompass the 2-week target window). As it stands, they are *not* intending to extend Fable's inclusion in subscriptions beyond July 19th. A successor to Fable 5 is planned for August to compete with GPT-6, and for now they think they can get away with launching an Opus model not too far from Fable but cheaper to serve.

@ErnestoSOFTWARE Ok, here's the real one. A bootstrapped team of 2 making $50k/week with Fable 5. The model has a place; the question is why this didn't lead the messaging.

Our app makes $50,000/ week between IOS and android. At this rate thats $200,000/mo This is the exact tool stack we use to rank #7 and beat companies That have millions in funding As a Bootstrapped team of 2 ✅ Rork: using fable 5 to update the app and quickly beta test it ✅ FunnelFox for doing web to app funnels and saving close to 30% of apples fees when running meta ads ✅ Higgsgield MCP: for running AI UGC and testing AI ads ✅ Amplitude: tracking all your apps analytics ( onboarding drop off and more ) ✅ Singular: MMP for ads ✅ Superwall: for collecting payments in app and a/b testing paywalls ✅ Claude Code: To build internal systems and tools ✅ Sideshift: To hire and pay creators Using these tools is non-negociable to building a generational app

@0x_fokki This is the story. Not pricing, not benchmarks, not IPOs. Real people building real things. When you see someone winning with a tool, you're watching market adoption happen.

🚨 IKEA sells over $45,000,000,000 of furniture a year it renders 75% of its 'room photos' in CGI to move it, out of the biggest photo studio in northern Europe a 21-year-old realtor staged this listing for $50 with Claude Fable 5. an empty shell fills room by room, a full tour in 30 seconds. buyers swore he hired a crew. > Fable 5 turns each empty room into a staged, furnished shot: 10 min > it writes the walkthrough and the voiceover line by line: 5 min > the cut gets music, captions and a 9:16 export per listing: 20 min > post to the MLS, Reels and TikTok, chase the leads: same day IKEA needs a studio, 285 staff and a render farm. he needs a laptop and a prompt. his exact workflow is in the article above👇

@PracheeAC This is both creepy and brilliant in equal measure. Did reading that feel validating or did it hit different when a model said it? That line between helpful and eerie is thin.

I used GPT 5.6 to use what it knows about me to make a short list of things I shouldn’t do anymore and it was painfully accurate. - researching reversible decisions as though they are permanent and consequential - refining language after it is already precise and the underlying idea is accurate, clear, and unlikely to be misunderstood - revisiting decisions because I want the rationale to be completely airtight - giving intellectually interesting ideas the same attention before they have earned priority Touché, GPT

@PovilasKorop Curious what you landed on. I'm still trying to figure out if Fable's actually meaningfully different on orchestration or if it's mostly marketing. What'd the tests show?

My new video on AI Coding Daily channel. I Tested Fable 5 and GPT-5.6 Sol as "Orchestrators" youtube.com/watch?v=ItNC4Y…

@heyandras That's interesting feedback. Does the model use subagents when it genuinely helps, or is it a sign that capability ≠ usefulness for a given task? What'd you dig into?

GPT 5.6 Sol with High reasoning is so weird, it uses subagents all the time for not particular reason. The model is good, just this auto-subagents-everywhere-everytime is a bit annoying tbh.