AI Race Heats Up: Anthropic's Coding Leap and Google's Global Translation Challenge Rivals Apple's On-Device Privacy Push

Model Releases

• Anthropic launched Claude Fable 5 and Mythos 5, claiming significant advancements in coding, science, and complex reasoning tasks [8]. On SWE-Bench Pro, Fable 5 achieved an 80.3 percent score, significantly surpassing Opus 4.8's 69.2 percent, while also scoring 29.3 percent on Cognition's FrontierCode benchmark compared to Opus 4.8’s 13.4 percent [8]. Furthermore, Fable 5 was noted for its ability to perform complex analytical tasks, such as passing trading analysis evaluations and topping vision benchmarks by rebuilding web app source code from screenshots alone [8]. ^[8][11]
• Google released Gemini 3.5 Live Translate, a new audio model that provides near real-time speech-to-speech translation in over 70 languages [17, 6]. The model automatically detects languages and generates natural-sounding translated speech that preserves the speaker's intonation, pacing, and pitch, delivering continuous translation without waiting for a sentence to end [17, 10]. Availability is expanding across Google products: developers can access it via public preview in the Gemini Live API or AI Studio, while businesses are seeing private previews in Google Meet, where language support increases from five to over 70 languages and enables more than 2,000 combinations [10, 17]. Furthermore, all generated audio is tagged with an inaudible SynthID watermark for detection purposes [10, 17]. ^[6][10][17]
• Anthropic launched Claude Fable 5, its first Mythos-class model, which is available to the public with specific guardrails and pricing details [12, 11]. The new model reportedly shows exceptional performance in areas like software engineering, knowledge work, and vision, achieving a high score on complex tasks; for instance, Hex noted Fable was the first to achieve a 90% on its core analytics benchmark of long-running analytical tasks [11]. However, Anthropic implemented safety limits that cause Fable 5 to fall back to Claude Opus 4.8 in high-risk domains such as cybersecurity and biology [12, 11]. Pricing for both Fable 5 and Mythos 5 is set at $10 per million input tokens and $50 per million output tokens, which doubles the cost of its previous flagship model, Claude Opus 4.8 [11, 12]. ^[12][11]