本周AI要点：Gemini 4 Argon 追平 GPT-6 Astra，Mistral Large 4 即将开源
一周模型动态都在这：Gemini 4 追平 GPT-6 还便宜四成，Mistral 放了个 1T 参数的开源模型，做选型或追前沿的可以扫一眼。
Google 的 Gemini 4 Argon 在 Artificial Analysis Intelligence Index 上拿下 53 分，与 GPT-6 Astra 持平，但单任务成本约为对方 60%。Mistral Large 4 拥有 1T 参数，开放权重版本将于 10 月下旬发布，成为美国和中国之外最聪明的模型。Reflection AI 的开源模型 Beam 推理所需算力比同级别开源模型少 3-4 倍。微软的 MAI-Transcribe-2-Streaming 在 38 个流式语音转文本系统中排名第一，Google 的 EmbeddingGemma 2 可在手机上运行多模态搜索。
Missed a few days of AI news? Here’s what happened in this week’s Data Points 🧵 🔹Google is back at the frontier. Gemini 4 Argon ties GPT-6 Astra at 53 on the Artificial Analysis Intelligence Index, at about 60% of the cost per task. 🔹Europe closes the gap. Mistral Large 4, a 1T-parameter model with open weights coming late October, is now the most intelligent model built outside the US and China. 🔹Reflection AI’s open-weight Beam claims 3–4x less compute to reason than comparable open models. 🔹OpenAI published a batch of new math results from an unreleased model, with proofs formalized in Lean. 🔹Microsoft’s MAI-Transcribe-2-Streaming ranks #1 of 38 for streaming speech-to-text accuracy. 🔹Google’s EmbeddingGemma 2 runs multimodal search on a phone; Cohere’s Embed 5 targets enterprise RAG. 🔹Meta’s Muse agent and your friends’ data, OpenAI’s opt-in text watermarks, arXiv’s new submission caps, and which AI subscriptions give the most value per dollar. Read it hubs.la/Q04ztBjc0 RaFlTN6HH 💬 0 🔄 1 ❤️ 5 👀 731 📊 1 ⚡
