九月模型盘点：Jev 被列为唯一不写句子的模型
逐条列出 9 月各家发布，把 TypeSafe AI 的 Jev 单列为异类：没有聊天窗口，只返回可分支的选择或概率，并给出价格与提速宣称。
Models -Claude Fable 5.1 and Mythos 5.1 (Sept 1): Anthropic's top tier for coding and knowledge work. @claudeai -GPT-6 Astra (Sept 3): OpenAI's flagship, and the model behind Dots. @OpenAI -Jev, TypeSafe AI (Sept 15): the odd one out. No chat window, no sentences. You send text and a question, it returns a choice or a probability your code can branch on. $0.042 per million input tokens, output free. The company claims 20 to 200x faster and 40 to 400x cheaper than frontier models. @CompleteSkeptic @typesafeai -Grok 4.7 (Sept 21): bigger base model at the same $2/$6 price. @xai -Claude Opus 5.5 (Sept 22): about Fable 5.1 quality at roughly 40% lower cost than Opus 5. @claudeai -GPT-6 Sol and Luna (Sept 22): mid and cheap tiers, 50% price cut against the 5.6 promo. @OpenAI -Claude Sonnet 5.5 (Sept 28): about 30% faster and up to 30% cheaper than Sonnet 5. @claudeai -GPT-6.1 Sol (Sept 29): OpenAI says near-Astra coding and computer use at about a fifth of the token cost. @OpenAI Every model was cheaper or faster, and one stopped writing sentences altogether. The models got cheaper, the agents multiplied, and the valuations left the products far behind. The launch that changes the most for brands is the quiet one. Agents can check out on Shopify now, and the biggest consumer agent has no ads inside it. Someone's assistant will soon do the shopping without watching a single ad, so brands will have to reach the person before they ask the assistant. Most of these agents won't exist in a year. The ones that do will hold your card.