AI 周报：Gemini 4 Argon 定价 $2/$10 每百万 token
周日 AI recap 提到 Google 的 Gemini 4 Argon 每百万 token 输入 2 美元、输出 10 美元，优惠持续到 11 月 30 日，另附 Meta、Anthropic 等动态。
you missed a lot in ai this week. here’s a quick sunday recap: - an ai model ran on fifteen $55 tv boxes. -four ai agents beat minecraft. - accenture capped claude at $29 a month per employee. here's everything that happened, what to try today, and what not to skip: the 3 you can't skip: - @Meta shipped muse wonder, its first frontier model ($4/$20 per M), and muse spark open (816B, open weights) - @AnthropicAI told investors q3 revenue hit $16.8B (per the ft) - claude code review is now GA on team and enterprise more models: - @Google : gemini 4 argon, $2/$10 per M until 30 nov - @Microsoft : mai-base-2, 2.8T params, closed - mistral large 4, @nvidia nemotron 3.5 super vl, grok voice 3.0 - opus 5.5 fast mode is open to everyone, no waitlist the weird ones: - anthropic ran a 26B model across fifteen $55 android tv boxes (0.39 tokens/sec) to test if ai could copy itself - four gemini agents beat minecraft in 27 hours. they died 605 times - xbow agents built a full chrome exploit in 6 days for ~$40k money and lawsuits: - accenture capped funded claude use at $29/employee/month. some were spending $1,000 - ftc and the eu ai office sent formal requests to openai and anthropic - 8 us states sued openai - 6 canadian publishers sued anthropic try this sunday (under 30 min each): - liquid d1-ultra: ask it a yes/no question about a photo. it answers with a probability, not text - janas: run a 48GB model on a 32GB laptop - opencode 2.0 with free deepseek v4.1 flash - claude code /code-review + a short REVIEW.md the pattern: generating code got cheap. reviewing, fixing and trusting it is where the work moved. which one are u trying first? bookmark this, i post one every sunday.