What changed in AI today

• 24B total parameters with 2B active per token, built as a reasoning model that thinks before answering.
• Scores 48.09 on LiveClaw versus 38.19 for Qwen3.6-35B-A3B, a model with roughly 50% more parameters.
• Decodes 184 tokens/s on an Apple M5 Max and 114 tokens/s on a Ryzen AI Max+ 395, per the source.
• Open-weight and available on Hugging Face now; supports llama.cpp, MLX, vLLM, SGLang, and ONNX from day one.

Why it matters for marketers: Content teams building tool-using assistants can run a local agent model, but should pair it with retrieval or search for knowledge-heavy tasks.

• Palabra.ai claims under 1 second latency and charges $0.04 per minute for API speech-to-speech translation.
• Google says Gemini 3.5 Live Translate covers 70+ languages but runs a few seconds behind the speaker.
• Apple's Live Translation runs entirely on device with 10 language variants, requiring recent AirPods and iOS 26.
• Teams Interpreter's language count is listed as 9 in admin docs and 10 on the support page.

Why it matters for marketers: Content teams planning multilingual calls or webinars should compare latency and per-minute cost, then pilot on their own language pairs, since counts aren't directly comparable.

• 1.6 billion parameter model; also transcribes English, French, Spanish and Portuguese with the same weights.
• Averages 20.92% WER across six Arabic test sets, versus 23.17% for the best published leaderboard result.
• Internal Emirati evaluation: 22.73% WER and 10.19% CER, 4.07 points below Qwen3-Omni, the next best.
• Live now as a Hugging Face Demo Space; API access and native applications are planned, not yet available.

Why it matters for marketers: Teams producing Arabic or Gulf-dialect transcripts or captions can test a free demo now, though API access is not yet released.