2. OpenAI & DeepMind dominate the world coding championship
At the 2025 ICPC World Finals, OpenAI’s GPT-5 reasoning system achieved a 12/12 in the AI track, outperforming all human teams. Google DeepMind’s Gemini 2.5 solved 10/12 problems, earning gold-level recognition
At the 2025 ICPC World Finals, OpenAI’s GPT-5 reasoning system achieved a 12/12 in the AI track, outperforming all human teams. Google DeepMind’s Gemini 2.5 solved 10/12 problems, earning gold-level recognition
❤2
3. China bans Nvidia AI chip purchases
China’s internet regulator ordered Alibaba, ByteDance, and others to halt
buying Nvidia’s newest AI chips, intensifying the U.S.–China tech war
China’s internet regulator ordered Alibaba, ByteDance, and others to halt
buying Nvidia’s newest AI chips, intensifying the U.S.–China tech war
❤2
4. Nvidia invests $5B in Intel
In a surprising move, Nvidia will co-develop PC and data center chips with Intel, integrating Nvidia GPUs into Intel CPUs. Intel stock jumped 26% in pre-market trading
In a surprising move, Nvidia will co-develop PC and data center chips with Intel, integrating Nvidia GPUs into Intel CPUs. Intel stock jumped 26% in pre-market trading
❤3
5. AI-designed virus kills harmful bacteria
Stanford and Arc Institute researchers created the first AI-generated virus that can successfully infect and kill bacteria, a milestone in computational biology
Stanford and Arc Institute researchers created the first AI-generated virus that can successfully infect and kill bacteria, a milestone in computational biology
❤3
6. Google brings Gemini AI to Chrome
Gemini is now built into Chrome for smarter browsing, with AI-powered search expanding to more global languages
Gemini is now built into Chrome for smarter browsing, with AI-powered search expanding to more global languages
❤3
7. OpenAI expands into hardware
OpenAI is hiring top Apple engineers and has struck a deal with iPhone assembler Luxshare to build future devices, bolstered by its acquisition of Jony Ive’s io Products
OpenAI is hiring top Apple engineers and has struck a deal with iPhone assembler Luxshare to build future devices, bolstered by its acquisition of Jony Ive’s io Products
❤2
8. DeepSeek reveals $294K training cost
Chinese startup DeepSeek disclosed that its R1 model training cost just $294,000, far cheaper than U.S. rivals spending hundreds of millions
Chinese startup DeepSeek disclosed that its R1 model training cost just $294,000, far cheaper than U.S. rivals spending hundreds of millions
❤3
BREAKING 🚨: OpenAI is planning to “throw more compute at new ideas” in coming weeks, exclusively for Pro subscribers and with additional fees.
Comet browser is now available to Perplexity Pro users in India. India became a second country after US to get broader access to Comet.
This media is not supported in your browser
VIEW IN TELEGRAM
Google released a new Mixboard experiment on @GoogleLabs in the US to let users play with a nano-banana-powered concepting board.
❤2🔥1
🚀 Alibaba floods market with Qwen3 upgrades
Six new models drop, led by 1T-param Qwen-Max.
Details:
• Max: near-frontier in coding + agent tasks
• Heavy: perfect math reasoning scores
• Omni: handles text, image, audio, video → 19 langs speech in, 10 out
• VL: top open-source visual model, beats many closed rivals
• Plus: LiveTranslate-Flash (real-time), Guard (safety), new Coders
Six new models drop, led by 1T-param Qwen-Max.
Details:
• Max: near-frontier in coding + agent tasks
• Heavy: perfect math reasoning scores
• Omni: handles text, image, audio, video → 19 langs speech in, 10 out
• VL: top open-source visual model, beats many closed rivals
• Plus: LiveTranslate-Flash (real-time), Guard (safety), new Coders
📊 Scale AI launches SEAL Showdown
New benchmark ranks LLMs by real user votes across demographics.
Details:
• Uses Scale’s global network → 100 countries, 70 languages
• Free access via Playground app → side-by-side model tests
• Voting optional + data locked 60 days to block gaming
• Leaderboards split by age, education, language.
https://scale.com/showdown
https://scale.com/leaderboard
New benchmark ranks LLMs by real user votes across demographics.
Details:
• Uses Scale’s global network → 100 countries, 70 languages
• Free access via Playground app → side-by-side model tests
• Voting optional + data locked 60 days to block gaming
• Leaderboards split by age, education, language.
https://scale.com/showdown
https://scale.com/leaderboard
OpenAI, Oracle, and SoftBank announced five data center sites across Texas, New Mexico, Ohio, and the Midwest for Stargate, pushing the project toward its 10GW target.
Suno released v5 of its music generation model, claiming new SOTA performance by a significant margin with new creative control and audio upgrades.
Microsoft published a cooling breakthrough in AI chips, etching tiny liquid channels to achieve 3x better heat removal and potentially solving AI’s “melting GPUs” problem.
Google Labs launched Mixboard, a new AI concept board that helps users visualize and refine ideas through text prompts and images with the Nano Banana editing model.
Abu Dhabi unveiled a new strategy to become the world's first fully AI-native government by 2027, planning to deploy 200+ AI solutions across the sector.
Suno released v5 of its music generation model, claiming new SOTA performance by a significant margin with new creative control and audio upgrades.
Microsoft published a cooling breakthrough in AI chips, etching tiny liquid channels to achieve 3x better heat removal and potentially solving AI’s “melting GPUs” problem.
Google Labs launched Mixboard, a new AI concept board that helps users visualize and refine ideas through text prompts and images with the Nano Banana editing model.
Abu Dhabi unveiled a new strategy to become the world's first fully AI-native government by 2027, planning to deploy 200+ AI solutions across the sector.
⚡ xAI unveils Grok 4 Fast
New reasoning model = near-frontier power at tiny cost.
Details:
• Uses 40% fewer thinking tokens → 98% cheaper
• Beats Claude 4.1 Opus + Gemini 2.5 Pro on key tests
• Scores: 85.7% GPQA Diamond, 92% AIME 2025
• #1 in LMArena Search Arena, top in coding (better than Grok 4)
• Supports 2M tokens + native web + code tools
New reasoning model = near-frontier power at tiny cost.
Details:
• Uses 40% fewer thinking tokens → 98% cheaper
• Beats Claude 4.1 Opus + Gemini 2.5 Pro on key tests
• Scores: 85.7% GPQA Diamond, 92% AIME 2025
• #1 in LMArena Search Arena, top in coding (better than Grok 4)
• Supports 2M tokens + native web + code tools
Trending AI Tools
🟥 Suno v5 - SOTA music generation model
🟩 Qwen3 Omni - Alibaba’s native end-to-end multilingual omni-modal LLM
🟥 Deepseek v3.1 Terminus - New model with upgraded agentic capabilities and performance
🟩 Qwen-Image-Edit - New multi-image editing and enhanced consistency
Please open Telegram to view this post
VIEW IN TELEGRAM
❤1❤🔥1
Please open Telegram to view this post
VIEW IN TELEGRAM
This media is not supported in your browser
VIEW IN TELEGRAM
Kimi AI released "OK Computer", a new agent mode which can build complete websites.
"An agentic model with its own computer, K2 now has true agency."
"An agentic model with its own computer, K2 now has true agency."
🤯4👍1
💻 Google Cloud DORA report: AI now mainstream in dev
90% of developers use AI daily, but trust remains low.
Details:
• Survey of 5,000 pros → avg 2h/day with AI assistants
• 30% trust outputs “a little” or “not at all” yet still use them
• 80% see higher productivity, 59% better code quality
• Google launched DORA AI Capabilities Model → 7 best practices
Why it matters: AI is core to dev work, but humans still double-check — adoption grows even as skepticism stays.
90% of developers use AI daily, but trust remains low.
Details:
• Survey of 5,000 pros → avg 2h/day with AI assistants
• 30% trust outputs “a little” or “not at all” yet still use them
• 80% see higher productivity, 59% better code quality
• Google launched DORA AI Capabilities Model → 7 best practices
Why it matters: AI is core to dev work, but humans still double-check — adoption grows even as skepticism stays.
🤝1