Claude Fable 5.1 is Ridiculous.

Summary This video, presented by the tech/gaming creator Cole, demonstrates using Anthropic's Claude Fable 5.1 model to generate playable 3D games from comprehensive text prompts and reference images. The presenter attempts to recreate three popular video games—EA Sports FC 27, Valorant, and Grand Theft Auto VI—evaluating the fidelity, game mechanics, and UI generated by the AI model.

What is shown

• [00:00] Intro highlighting Anthropic's release of Claude Fable 5.1 and Claude Mythos 5.1, showing a benchmark table comparing Fable 5.1 against Fable 5, Opus 5, and GPT-5.6 Sol.
• [00:18] Claude interface showing model selection set to "Fable 5.1" with effort dialed to "Max".
• [00:26] Presenting the prompt and reference images used to generate FC 27 (UI menus, stadium views, gameplay).
• [00:50] Playtesting the generated FC 27 clone ("45 minutes later"), featuring menus, kickoff match mode, animated player models, ball physics, passing, camera switching via the "V" key, and goal scoring animations.
• [02:14] Crafting and submitting an extensive prompt with reference UI, map overview, buy phase, and gameplay screenshots to recreate Valorant.
• [02:44] Playtesting the generated Valorant recreation ("1 hour later"), including the main menu UI, match loading screen with agent cards, buy phase interface, weapon purchasing, abilities (dash), and 3D first-person shooter combat.
• [05:31] Inputting a prompt and screenshots of Grand Theft Auto VI (Vice City / Leonida) gameplay, cutscenes, driving, and mini-map.
• [05:55] Playtesting the GTA VI ("Leonida VI") recreation ("2 hours later"), showing a city cutscene, character controls, an NPC mission conversation with Lucia, driving physics across city streets and bridges, a full pause menu map, switching cars, and weapon aiming.
• [08:54] Mention of OpenAI's recently released Astra 6 (GPT-6 Astra) model as potential competition.

Claims & numbers

• The benchmark graphic claims Claude Fable 5.1 scores 52.6% on Agentic Scientific Research (Terminal-Bench-Science 0.1), 55.8% on Agentic Coding (Terminal-Bench 4.0, with Mythos 5.1 at 60.9%), 1853 on Knowledge Work (GDPval-AA v2), 77.9% partial / 41.7% strict on OSWorld 2.0 Computer Use, 60.9% no tools / 65.0% with tools on Humanity's Last Exam Multidisciplinary Reasoning, 21.4% on Business Workflows AutomationBench, and 73.4% on Agentic Coding Cursortest-Bench 2.0 [00:04].
• The presenter claims the benchmarks show Fable 5.1 is "the best model by far" [00:04].
• Generating the FC 27 game took approximately 45 minutes of processing time [00:50].
• Generating the Valorant game took approximately 1 hour of processing time [02:42].
• Generating the GTA VI recreation took approximately 2 hours and consumed all of the presenter's Fable credits [05:55].
• The presenter notes that OpenAI recently released their "Astra 6" (GPT-6 Astra) model, which some claim outperforms Fable 5.1 [08:54].

Notable quotes

• [02:19] "I just sent in this prompt, and this might be my best prompt ever."
• [03:44] "This is genuinely light years better, and it's like the next model. So when Claude drops Fable 5.2, it's actually over for humanity."
• [08:34] "You can tell how Fable 5.1 is just light years ahead of all the previous AIs."

Assessment This is a creator review and demonstration video testing the game-generation coding and multimodal capabilities of Claude Fable 5.1. While the gameplay demonstrations showcase working 3D web/engine environments generated from prompts, the process involves significant generation wait times (45 minutes to 2 hours) and clearly uses pre-made low-poly 3D asset packs and template game frameworks orchestrated via Claude rather than generating AAA-fidelity code from scratch.

Described by gemini-3.8-flash on 2026-09-29 from the video's audio and frames.