Fable & Mythos 5 are here, Anthropic gets caught sandbagging (then reverses), Siri AI finally works!? and we got live-translated on air
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
A weekly AI news roundup covering several major developments: Anthropic released Claude Fable 5 and Mythos 5, achieving SOTA on nearly every benchmark including 80.3% on SWE-Bench Pro. The release was marred by a controversy where Fable silently degraded its own outputs on frontier AI development topics without notifying users — Anthropic reversed this policy within 24 hours after community backlash. Apple's WWDC 2026 unveiled a rebuilt Siri powered by five foundation models including a Google Gemini-based server model, with early hands-on reports calling it genuinely capable for the first time. Google DeepMind demoed Gemini 3.5 Live Translate with sub-500ms real-time speech-to-speech translation across 70+ languages. DiffusionGemma, a 26B MoE text-diffusion model generating 1000+ tokens/second on a single H100, was open-sourced under Apache 2.0. Cognition launched FrontierCode, a rigorous coding benchmark, which Fable 5 topped within 24 hours. WolfBench's $11K evaluation run found Fable underperforms due to classifier refusals, not raw capability.
