01-13-Daily AI News Daily

Apple and Google are officially teaming up, with Siri getting a Gemini brain. This means billions of device users are now Google's new "chives" (a playful term for users who are easily harvested).
Claude has rolled out Cowork, letting regular folks command AI to organize files and create spreadsheets. Meanwhile, the large model throne is a hot seat, lasting only 35 days on average, with o1 already plummeting to 56th place.
Today's news makes one thing crystal clear: building it yourself often pales in comparison to riding on someone else's coattails, and "wait-and-see" folks are always right.

⚡ Quick Navigation

📰 Today’s AI News - A quick glance at the latest happenings.

💡 Tip

Want to be among the first to try out the latest AI models mentioned (Claude 4.5, GPT, Gemini 3 Pro)? No account? Head over to Aivora to grab an account, get started in a minute, and enjoy worry-free after-sales support.

👀 Just One Sentence

Apple officially announced its partnership with Google, meaning Siri is getting Gemini’s brain.

🔑 3 Keywords

#AppleGoogleMarriage #ClaudeGoesMainstream #ModelThrone35DayRotation

🔥 Top 10 Heavy Hitters

  1. Apple Announces Partnership with Google, Gemini to Power Apple Intelligence

    Apple and Google, those “old rivals,” are actually teaming up – you heard that right! Apple’s official statement is pretty blunt: “After careful evaluation, we’ve determined that Google’s technology provides the strongest foundation for Apple’s intelligent models.” In plain English, that translates to: “We couldn’t quite nail it ourselves, so we’re using yours.” Siri will be powered by the Gemini model in the future, with a more personalized version expected later this year. For Google, this is a no-brainer win-win: billions of Apple devices are about to become Gemini users. Google stock? Nuff said.

  2. Claude Launches Cowork: AI Can Now Help You Work Even Without Coding Skills

    Claude Code was initially for developers, but folks soon realized it was also super handy for organizing files, making spreadsheets, and writing reports. So, Anthropic smartly packaged this capability as Cowork, making it accessible to everyone. Cowork lets you pick a folder on your computer, and Claude can then read, modify, and create new files within it. Is your downloads folder a hot mess? Let it sort and rename everything for you. Got a bunch of expense screenshots? It can whip them into an Excel sheet. Unlike regular chat, Cowork mode makes Claude feel more like a genuine assistant – you set the task, and it plans and executes the steps itself. Currently, it’s only available to Claude Max subscribers on Mac, with a Windows version on the way.

  3. Large Model Throne Lasts Only 35 Days on Average, o1 Drops to 56th Place

    lmarena crunched the numbers on all models that have topped the charts since mid-2023, and the results are a bit brutal: starting with GPT-4, models only hold the #1 spot for about 35 days on average. We’re not talking half a year or even a full year; a new model overtakes them in just over a month. What’s even wilder is that many models fall out of the Top 5 around five months after peaking, and struggle to stay in the Top 10 after seven months. The once-glorious o1 is now ranked #56, and Claude 3 Opus, once hailed as the “strongest reasoning model,” has slipped to #139. What does this tell us? The pace of fundamental model capability improvement is outstripping the iteration cycles of most products.

  4. Yuan 3.0 Flash: 40B Parameters, 75% Reduction in Inference Token Usage

    YuanLab.ai team has unleashed Yuan 3.0 Flash, a large model boasting 40B parameters. Its standout feature is the MOE architecture, requiring only 3.7B active parameters per inference, matching GPT-5.1 in tests. It can go head-to-head with 235B models in inference accuracy, but with a whopping 75% reduction in token usage. For teams tight on computing power, this is fantastic news – achieving better results with fewer resources. The model is already live on ModelScope, so if you’re curious, go give it a whirl.

  5. DeepSeek Releases Engram: Enhancing Model Contextual Memory

    DeepSeek is at it again, dropping something big late at night. This time it’s Engram, a fresh approach to conditional memory via scalable lookups, adding a new dimension of sparsity to large language models. Engram, simply put, helps models “remember” key information better when processing long texts, instead of getting forgetful as they go along. This is a significant breakthrough for scenarios requiring the handling of ultra-long documents. The code is already open-source on GitHub.

  6. Gemini Image Generation Surpasses 1 Billion Images, Nano Banana Pro Model Only 53 Days Old

    Google’s Josh Woodward announced that the Nano Banana Pro image generation model in the Gemini App has already surpassed 1 billion images. This model has only been live for 53 days, and the Pro community’s creation speed is absolutely mind-blowing. What does 1 billion images even mean? It’s nearly 20 million images generated daily on average. This clearly shows that AI image generation has evolved from a “novelty” to an everyday tool, boasting incredibly high user stickiness.

  7. Veo 3.1’s Start/End Frame Feature is Insane: Inserting Animation Between Two Images

    Veo 3.1’s start/end frame feature is a real eye-opener. Someone tested it with a vector illustration-style PPT, and it actually auto-generated a vector animation of a car driving between two static images, with a natural and smooth transition. What does this mean? For future PPT presentations and short video transitions, you might just need to give AI two images, and it can fill in the animation in between. Designers and video creators are about to see their efficiency jump another level.

  8. ByteDance and Alibaba Prepare to Spend 75 Billion to Snatch NVIDIA H200 GPUs

    Foreign media reports indicate that China will soon permit the purchase of NVIDIA H200 GPUs in the first quarter. ByteDance and Alibaba have already privately expressed their intent to NVIDIA, with each aiming to order 200,000 H200 cards. Let’s do the math: 200,000 cards translate to 25,000 8-GPU servers. Each company’s purchase amounts to 37.5 billion, totaling 75 billion RMB for both. Plus, NVIDIA’s terms are pretty hardcore: 100% upfront payment, no cancellations, no refunds. But hey, you gotta do what you gotta do – this is buying the “privilege to burn compute.” With these 400,000 cards, companies can shift from mere engineering optimization to genuine technological exploration.

  9. opencode Integrates GitHub Copilot, API Freedom Achieved

    opencode now supports selecting GitHub Copilot as a model service. If your company has an enterprise Copilot subscription, it’s essentially API freedom – use top-tier models to your heart’s content, and if you blow through your quota, the company foots the bill. Users who’ve tested it found it super smooth for prototyping and coding, with some not having manually drawn a prototype in over half a year. For teams with an enterprise Copilot subscription, this is a golden opportunity to get some serious freebies.

  10. planning-with-files Open Source: 7000 Stars in 3 Days, Replicating Manus Workflow

    The open-source project called planning-with-files has gone viral, with its core concept being “planning with files.” After installing this Skill, Claude automatically maintains three core files: task_plan.md (task progress), notes.md (research materials), and deliverable.md (final output). Basically, this is like giving AI an external brain plugin. By forcing AI to use local files to record progress and thoughts, it solves the pain point of large models forgetting things during long conversations. Any developer working on Agents absolutely needs to check this out.

📌 Worth Noting

😄 AI Fun Facts

Goubuli Baozi Adopts English Name Go Believe

Today’s most outlandish AI-related news: the venerable Chinese brand Goubuli Baozi’s English name, “GO BELIEVE,” has sparked a buzz. Netizens find the translation clever due to its phonetic similarity to “Goubuli.” Tianyancha shows that Goubuli Group registered this trademark way back in 2007. Store staff say the name was chosen to make it easier to introduce to foreign guests. Netizen comments: “Even AI couldn’t come up with a translation this good!”

🔮 AI Trend Predictions

  • Apple Siri Major Upgrade to Debut at WWDC 2026

    Prediction Time: June 2026 Prediction Probability: 75% Prediction Basis: Today’s news Apple Announces Partnership with Google + Apple’s statement mentioning “a more personalized Siri coming later this year,” making WWDC the perfect launchpad.

  • Qwen and Doubao to Release Major Updates in Q2

    Prediction Time: Q2 2026 Prediction Probability: 70% Prediction Basis: Today’s news ByteDance and Alibaba Prepare to Spend 75 Billion to Snatch H200 + large-scale compute purchases usually signal model breakthroughs 3-6 months later.

  • Claude Cowork to Open Windows Version in Q1

    Prediction Time: Q1 2026 Prediction Probability: 65% Prediction Basis: Today’s news Claude Launches Cowork + Anthropic explicitly stating “a Windows version will follow,” and research previews typically expand scope after 1-2 months.

❓ Related Questions

How to Experience Claude Cowork?

Claude Cowork is currently in research preview, exclusively available to Claude Max subscribers on Mac. A Claude Max subscription requires an overseas payment method, which might pose payment difficulties or account registration restrictions for users in mainland China. Solution: Visit Aivora to get a ready-made account, with lightning-fast delivery and worry-free after-sales support.

Today’s Digest 

Apple and Google are officially teaming up, with Siri getting a Gemini brain. This means billions of device users are now Google's new "chives" (a playful term for users who are easily harvested).
Claude has rolled out Cowork, letting regular folks command AI to organize files and create spreadsheets. Meanwhile, the large model throne is a hot seat, lasting only 35 days on average, with o1 already plummeting to 56th place.
Today's news makes one thing crystal clear: building it yourself often pales in comparison to riding on someone else's coattails, and "wait-and-see" folks are always right.

⚡ Quick Navigation

💡 Tip: Want to be among the first to try out the latest AI models mentioned (Claude 4.5, GPT, Gemini 3 Pro)? No account? Head over to Aivora to grab an account, get started in a minute, and enjoy worry-free after-sales support.

Today’s AI News

👀 Just One Sentence

Apple officially announced its partnership with Google, meaning Siri is getting Gemini’s brain.

🔑 3 Keywords

#AppleGoogleMarriage #ClaudeGoesMainstream #ModelThrone35DayRotation

🔥 Top 10 Heavy Hitters

1. Apple Announces Partnership with Google, Gemini to Power Apple Intelligence

Apple and Google, those “old rivals,” are actually teaming up – you heard that right! Apple’s official statement is pretty blunt: “After careful evaluation, we’ve determined that Google’s technology provides the strongest foundation for Apple’s intelligent models.” In plain English, that translates to: “We couldn’t quite nail it ourselves, so we’re using yours.”

Siri will be powered by the Gemini model in the future, with a more personalized version expected later this year. For Google, this is a no-brainer win-win: billions of Apple devices are about to become Gemini users. Google stock? Nuff said.

Image

2. Claude Launches Cowork: AI Can Now Help You Work Even Without Coding Skills

Claude Code was initially for developers, but folks soon realized it was also super handy for organizing files, making spreadsheets, and writing reports. So, Anthropic smartly packaged this capability as Cowork, making it accessible to everyone.

Cowork lets you pick a folder on your computer, and Claude can then read, modify, and create new files within it. Is your downloads folder a hot mess? Let it sort and rename everything for you. Got a bunch of expense screenshots? It can whip them into an Excel sheet. Unlike regular chat, Cowork mode makes Claude feel more like a genuine assistant – you set the task, and it plans and executes the steps itself. Currently, it’s only available to Claude Max subscribers on Mac, with a Windows version on the way.

3. Large Model Throne Lasts Only 35 Days on Average, o1 Drops to 56th Place

lmarena crunched the numbers on all models that have topped the charts since mid-2023, and the results are a bit brutal: starting with GPT-4, models only hold the #1 spot for about 35 days on average. We’re not talking half a year or even a full year; a new model overtakes them in just over a month.

What’s even wilder is that many models fall out of the Top 5 around five months after peaking, and struggle to stay in the Top 10 after seven months. The once-glorious o1 is now ranked #56, and Claude 3 Opus, once hailed as the “strongest reasoning model,” has slipped to #139. What does this tell us? The pace of fundamental model capability improvement is outstripping the iteration cycles of most products.

4. Yuan 3.0 Flash: 40B Parameters, 75% Reduction in Inference Token Usage

YuanLab.ai team has unleashed Yuan 3.0 Flash, a large model boasting 40B parameters. Its standout feature is the MOE architecture, requiring only 3.7B active parameters per inference, matching GPT-5.1 in tests. It can go head-to-head with 235B models in inference accuracy, but with a whopping 75% reduction in token usage.

For teams tight on computing power, this is fantastic news – achieving better results with fewer resources. The model is already live on ModelScope, so if you’re curious, go give it a whirl.

Image

5. DeepSeek Releases Engram: Enhancing Model Contextual Memory

DeepSeek is at it again, dropping something big late at night. This time it’s Engram, a fresh approach to conditional memory via scalable lookups, adding a new dimension of sparsity to large language models.

Engram, simply put, helps models “remember” key information better when processing long texts, instead of getting forgetful as they go along. This is a significant breakthrough for scenarios requiring the handling of ultra-long documents. The code is already open-source on GitHub.

Image

6. Gemini Image Generation Surpasses 1 Billion Images, Nano Banana Pro Model Only 53 Days Old

Google’s Josh Woodward announced that the Nano Banana Pro image generation model in the Gemini App has already surpassed 1 billion images. This model has only been live for 53 days, and the Pro community’s creation speed is absolutely mind-blowing.

What does 1 billion images even mean? It’s nearly 20 million images generated daily on average. This clearly shows that AI image generation has evolved from a “novelty” to an everyday tool, boasting incredibly high user stickiness.

7. Veo 3.1’s Start/End Frame Feature is Insane: Inserting Animation Between Two Images

Veo 3.1’s start/end frame feature is a real eye-opener. Someone tested it with a vector illustration-style PPT, and it actually auto-generated a vector animation of a car driving between two static images, with a natural and smooth transition.

What does this mean? For future PPT presentations and short video transitions, you might just need to give AI two images, and it can fill in the animation in between. Designers and video creators are about to see their efficiency jump another level.

8. ByteDance and Alibaba Prepare to Spend 75 Billion to Snatch NVIDIA H200 GPUs

Foreign media reports indicate that China will soon permit the purchase of NVIDIA H200 GPUs in the first quarter. ByteDance and Alibaba have already privately expressed their intent to NVIDIA, with each aiming to order 200,000 H200 cards.

Let’s do the math: 200,000 cards translate to 25,000 8-GPU servers. Each company’s purchase amounts to 37.5 billion, totaling 75 billion RMB for both. Plus, NVIDIA’s terms are pretty hardcore: 100% upfront payment, no cancellations, no refunds. But hey, you gotta do what you gotta do – this is buying the “privilege to burn compute.” With these 400,000 cards, companies can shift from mere engineering optimization to genuine technological exploration.

Image

9. opencode Integrates GitHub Copilot, API Freedom Achieved

opencode now supports selecting GitHub Copilot as a model service. If your company has an enterprise Copilot subscription, it’s essentially API freedom – use top-tier models to your heart’s content, and if you blow through your quota, the company foots the bill.

Users who’ve tested it found it super smooth for prototyping and coding, with some not having manually drawn a prototype in over half a year. For teams with an enterprise Copilot subscription, this is a golden opportunity to get some serious freebies.

Image

10. planning-with-files Open Source: 7000 Stars in 3 Days, Replicating Manus Workflow

The open-source project called planning-with-files has gone viral, with its core concept being “planning with files.” After installing this Skill, Claude automatically maintains three core files: task_plan.md (task progress), notes.md (research materials), and deliverable.md (final output).

Basically, this is like giving AI an external brain plugin. By forcing AI to use local files to record progress and thoughts, it solves the pain point of large models forgetting things during long conversations. Any developer working on Agents absolutely needs to check this out.

Image

📌 Worth Noting

😄 AI Fun Facts

Goubuli Baozi Adopts English Name Go Believe

Today’s most outlandish AI-related news: the venerable Chinese brand Goubuli Baozi’s English name, “GO BELIEVE,” has sparked a buzz. Netizens find the translation clever due to its phonetic similarity to “Goubuli.” Tianyancha shows that Goubuli Group registered this trademark way back in 2007. Store staff say the name was chosen to make it easier to introduce to foreign guests. Netizen comments: “Even AI couldn’t come up with a translation this good!”

Image

🔮 AI Trend Predictions

Apple Siri Major Upgrade to Debut at WWDC 2026

  • Prediction Time: June 2026
  • Prediction Probability: 75%
  • Prediction Basis: Today’s news Apple Announces Partnership with Google + Apple’s statement mentioning “a more personalized Siri coming later this year,” making WWDC the perfect launchpad.

Qwen and Doubao to Release Major Updates in Q2

Claude Cowork to Open Windows Version in Q1

  • Prediction Time: Q1 2026
  • Prediction Probability: 65%
  • Prediction Basis: Today’s news Claude Launches Cowork + Anthropic explicitly stating “a Windows version will follow,” and research previews typically expand scope after 1-2 months.

❓ Related Questions

How to Experience Claude Cowork?

Claude Cowork is currently in research preview, exclusively available to Claude Max subscribers on Mac. A Claude Max subscription requires an overseas payment method, which might pose payment difficulties or account registration restrictions for users in mainland China.

Solution: Visit [Aivora

Last updated on
01-14-Daily01-12-Daily