AI NEWS: NEW Fable 5.2” and “Gemini 4 Pro” leaks!?
People spent the weekend saying Claude Fable 5.1 is quietly routing to an unreleased "Fable 5.2", and that Google is secretly testing "Gemini 4 Pro". I went through every post, ran the test everyone is running, and checked both companies' own documentation. Here is what is real, what is a label somebody typed themselves — and two games you can go play right now. 🔔 SUBSCRIBE TO BYTEFORWARD AI news, coding agents, model comparisons, benchmarks, and the things people are actually building: https://www.youtube.com/channel/UCGGHPOAzUaKsSc2yjsOeAFQ?sub_confirmation=1 ▶ PLAY THESE YOURSELF (both by @TaylorBereiter, run in a browser, no install) Voxnite Battle Royale — https://claude.ai/artifact/9mGKuM7NiH69rZouD8QHdh (desktop: it needs a keyboard and mouse) Inkwash Keep — https://claude.ai/artifact/4AvaiZmqh6u8zuHbARHBXf ▶ THE POSTS IN THIS VIDEO Voxnite — @TaylorBereiter: https://x.com/TaylorBereiter/status/2101328500074639623 Voxnite quote-post — @notjazii: https://x.com/notjazii/status/2101334135507910663 Sundown Showdown — @notjazii: https://x.com/notjazii/status/2100981677950730388 ...and his own reply crediting it: https://x.com/notjazii/status/2100983868304269727 Inkwash Keep — @TaylorBereiter: https://x.com/TaylorBereiter/status/2101359786495516713 Ballpoint-pen game and trailer — @chetaslua: https://x.com/chetaslua/status/2101071418507366895 Front-end page — @Haleeeemahh: https://x.com/Haleeeemahh/status/2101029299717320986 Rocket test, both labels — @Bhavani_00007: https://x.com/Bhavani_00007/status/2100974061392478230 "Stealth testing" post — @clawdb0t: https://x.com/clawdb0t/status/2100905642752122992 ...the post it rewrites — @notjazii: https://x.com/notjazii/status/2100877638634189238 ...and the quoted video, which is @Ananth7e's: https://x.com/clawdb0t/status/2099883043091812477 ▶ WHERE THE "FABLE 5.2" CLAIM STARTED @notjazii, Sep 18: https://x.com/notjazii/status/2100893667846402306 @notjazii, Sep 14 (the first "Tibo" test, aimed at Opus 5): https://x.com/notjazii/status/2099498646278688981 @notjazii, the same day: "name is not confirmed so i assumed it's gonna be fable 5.2" — https://x.com/notjazii/status/2100941084025290762 The same claim one model earlier, @legit_api on Aug 26: https://x.com/legit_api/status/2092633561119482037 ▶ THE "GEMINI 4 PRO" POSTS Pelican on a bicycle — @LuminaBench: https://x.com/LuminaBench/status/2100575220947964003 PS5 controller — @LuminaBench: https://x.com/LuminaBench/status/2100590415141327080 IWC watch, three models — @YouWareAI: https://x.com/YouWareAI/status/2100956517633761447 Mechanical butterfly — @thtbee_: https://x.com/thtbee_/status/2101304547876700627 The article going round — Atoms: https://atoms.dev/blog/gemini-4-pro-examples ▶ WHO TIBO IS, AND WHY THE TEST DOESN'T WORK Tibo (@thsottiaux) leads Codex at OpenAI. His first public reset post, Sep 17 2025: https://x.com/thsottiaux/status/1968163721034994139 April 28 2026, 16,927 likes: https://x.com/thsottiaux/status/2048997818673537399 The reset tracker (53 logged, 27 of them before Fable 5.1's cutoff): https://codex-resets.com/ ▶ WHAT THE COMPANIES ACTUALLY SAY Anthropic model docs — Fable 5.1, knowledge cutoff June 2026: https://platform.claude.com/docs/en/about-claude/models/overview Anthropic — Claude Fable 5.1 launch, Sep 1 2026: https://www.anthropic.com/claude-fable-and-mythos-5-1 Google — "our most ambitious pre-training run yet, for Gemini 4" (Alphabet Q2 2026 earnings): https://blog.google/company-news/inside-google/message-ceo/alphabet-earnings-q2-2026/ Google DeepMind — the next Pro model is Gemini 3.5 Pro, "coming soon": https://deepmind.google/models/gemini/ Arena text leaderboard, checked Sep 19 2026: https://arena.ai/leaderboard/text As of recording, neither Anthropic nor Google has announced or confirmed any of this. Every "5.2" and every "Gemini 4 Pro" you see on screen in this video was typed there by the person posting it. All clips play at native speed and are credited to the people who made them. ▶ CHAPTERS 0:00 The game that started this 0:35 Where the “Fable 5.2” claim came from 1:06 The “Tibo the reset guy” test 1:36 Why that test proves almost nothing 2:28 The outputs, though 2:33 Voxnite: one prompt, 77 minutes 3:11 Sundown Showdown, and who actually made it 3:31 The credit that contradicts the post 4:17 The repost layer 4:42 Google’s half: “Gemini 4 Pro” 5:23 What Google has actually said 5:43 The “#1 on Arena” claim 6:01 The article doing the rounds 6:23 The precedent that keeps me honest 7:05 Play them yourself — both links 7:25 See you in the next one #AI #Claude #Anthropic #Gemini #GoogleDeepMind #AInews #ByteForward
