Latest in AI · updated Oct 21, 2026
The AI news that matters for you and your bots, in plain words. Top story = most important. Trending = most likes and reposts on X right now. Every story links to its source.
30 stories · 🧠 model release: 14 · 🤖 agent launch: 3 · ✨ feature: 6 · 🏷 pricing: 4 · 🏢 industry: 3 · 25 with measured X numbers · RSS feed · Price changes also go into the AI price table. How we measure.
For you: Claude Duet sends you a morning brief from your connected apps, keeps an eye on them through the day, works on tasks while you're away, and asks before it does anything big like paying or deleting. It's a waitlist research preview for US Pro and Max subscribers, on iOS, Android and claude.ai, and Anthropic says it can still misjudge what's urgent.
For you: DeepInfer is the software DeepSeek uses to serve V4.1 Flash cheaply, now free under Apache 2.0.
Source: DeepSeek (GitHub)
For you: If you run a website or blog on PHP (WordPress included), make sure your host has updated PHP to the patched version now.
Source: OpenAI
For you: DeepSeek says the update finishes tasks faster with shorter outputs, which means fewer tokens to pay for.
Source: DeepSeek API changelog
For you: Teams on Claude Enterprise can connect their code once and Claude keeps checking new changes for security holes, marking each one fixed when the fix lands.
Source: Anthropic
For you: Companies on Google Cloud will be able to use GPT-6 models and Codex without a separate OpenAI account (private preview now, wider launch planned in the coming weeks).
For you: A new set of speech models plus a kit for building phone and web voice agents that listen, look things up and talk back in 50+ languages.
Source: Mistral
For you: Bots that run hours apart (like a morning and an evening check) can now keep their fixed instructions cached and pay the lower cached-input price.
Source: OpenAI API changelog
💸 What this means for your bill →For you: From November 17, Google starts converting personal Gems into Skills you call with a slash in the main chat, and Scheduled Actions no longer need a paid plan (still rolling out to free accounts).
Source: Android Police
For you: Bots built on DeepSeek can now search the web and read pages through DeepSeek's own API at $0.75 per 1,000 requests each.
Source: DeepSeek pricing
💸 What this means for your bill →For you: Meta's new open model is built for coding and agent work, and you can call it on day one through Groq, OpenRouter, Together AI and other APIs.
Source: Meta · Also: Groq (limited preview), Hugging Face
For you: The voice model behind Grok Bot's voice mode and Tesla's in-car assistant got an upgrade, at the same $0.08 a minute on the API.
Source: xAI · Also: SiliconANGLE
For you: Google's new top model is in the API at an introductory $2 / $10 per million tokens, and in the Gemini app for Google AI Pro and Ultra subscribers.
Source: Google · Also: Gemini API release notes
For you: Meta says Muse Wonder will soon power its Muse personal agent, starting with paying subscribers.
Source: Meta · Also: TechCrunch
For you: Handing a job to Google's background agent is now one tap away in the Gemini app, with a new Scheduled page that shows active, paused and finished tasks.
Source: 9to5Google · Also: Android Authority
For you: Agents built on OpenAI's Agents API run in a hosted sandbox computer, and that is now billed only during active turns, not while the agent waits for you.
Source: OpenAI API changelog
💸 What this means for your bill →For you: Ask a question in plain words and Claude builds a dashboard that stays current, showing the query behind every number.
Source: Anthropic · Also: Reuters via Investing.com
For you: ChatGPT answers can now include charts, buttons, forms and small tools you use right in the chat.
Source: OpenAI
For you: At $0.10 / $0.50 per million tokens for prompts up to 100,000 tokens, Anthropic says it costs about 75% less to run than Haiku 4.5.
Source: Anthropic
For you: Cached reads on Claude Sonnet 5.5 dropped from $0.20 to $0.10 per million tokens, which Anthropic says makes most agent work about 20% cheaper.
Source: Anthropic
💸 What this means for your bill →For you: Mistral's biggest model yet is in a public preview on its API, and Mistral says anyone will be able to download and run it themselves later this month.
Source: Mistral
For you: A Claude sidebar can read the file you have open and edit it in place, asking before each change by default.
For you: OpenAI says GPT-6.1 Sol nearly matches its top model, GPT-6 Astra, on agent work at one-fifth of Astra's token prices ($2 / $10 per million).
Source: OpenAI · Also: OpenAI API changelog
For you: Agents built on OpenAI's platform can now complete tasks in an OpenAI-hosted browser, with the app asking you before it visits a site and handling sign-in.
Source: OpenAI API changelog
For you: A dot runs on GPT-6 Astra with its own cloud computer, connects to your apps, and keeps working between chats, checking with you on decisions.
Source: OpenAI · Also: OpenAI Help Center, The Verge
For you: Anthropic says Sonnet 5.5 runs 30%+ faster than Sonnet 5 and costs up to 30% less for most work.
Source: Anthropic
For you: One person sets up a bot with the files, apps and skills it needs, then the whole team can use it in Slack or Grok Bot, while each person's chats stay private.
Source: xAI · Also: TeslaNorth
For you: Anthropic says Opus 5.5 leads in agentic coding and knowledge work and costs 40% less to run than Opus 5 on typical workloads.
Source: Anthropic
For you: Two new models: Sol for most work and Luna, a very cheap one at $0.10 / $0.50 per million tokens.
Source: OpenAI API changelog
For you: xAI's new model for coding and knowledge work keeps the same $2 / $6 per million token price as Grok 4.6.
Source: xAI · Also: SiliconANGLE
Nothing in this filter yet.
How we rank. Heat = (likes + 2 × reposts) on the announcing X post (or the most-engaged post about it), divided by weeks since it was posted. Numbers were measured on the date in each tooltip. Stories with no post we could find get no number. Major means a new flagship model or a big agent launch from OpenAI, Anthropic, Google, Meta or xAI. It describes importance, not popularity.