{"slug": "fable-5-pushed-gemma-4-to-255-tok-s-on-webgpu", "title": "Fable 5 pushed Gemma 4 to 255 tok/s on WebGPU", "summary": "Fable 5, an AI agent, achieved 255 tokens per second on Gemma 4 inference using WebGPU before its access was suspended. The developer released the demo and kernels, claiming agentic kernel optimization is the future of on-device inference.", "body_md": "Before Fable 5 was shut down, it pushed Gemma 4 to 255 tok/s on WebGPU. Some didn't believe it was real.\nToday we're releasing the demo and kernels it wrote for you to see yourself. Run it locally in your browser.\nAgentic kernel optimization is the future of on-device inference\n\nI gave Fable 5 one job: write custom WebGPU kernels for Gemma 4 inference.\nIt climbed to 84 tok/s, then hit a wall, insisting further optimization was impossible.\nHours later, Anthropic rolled back invisible LLM development safeguards, and it hit 255 tok/s.\nThe next day, access to Fable 5 was suspended globally.\n\nJun 17, 2026 · 4:54 PM UTC\n\n69\n\n160\n\n1,733\n\n264,359\n\nIn case you hadn't noticed, we're working on something big. Stay tuned.\n🔗 Link to the demo:\n\n[huggingface.co/spaces/webml-…](https://huggingface.co/spaces/webml-community/gemma-4-webgpu-kernels) 5\n\n9\n\n120\n\n7,879", "url": "https://wpnews.pro/news/fable-5-pushed-gemma-4-to-255-tok-s-on-webgpu", "canonical_source": "https://xcancel.com/xenovacom/status/2067289897111638484", "published_at": "2026-06-18 14:14:58+00:00", "updated_at": "2026-06-18 14:22:57.942498+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-agents", "ai-research", "ai-tools", "ai-infrastructure"], "entities": ["Fable 5", "Gemma 4", "WebGPU", "Anthropic"], "alternates": {"html": "https://wpnews.pro/news/fable-5-pushed-gemma-4-to-255-tok-s-on-webgpu", "markdown": "https://wpnews.pro/news/fable-5-pushed-gemma-4-to-255-tok-s-on-webgpu.md", "text": "https://wpnews.pro/news/fable-5-pushed-gemma-4-to-255-tok-s-on-webgpu.txt", "jsonld": "https://wpnews.pro/news/fable-5-pushed-gemma-4-to-255-tok-s-on-webgpu.jsonld"}}