🧑💻 New Chinese Model Codes As Well As Claude 4 Opus and GPT‑4.1
Chinese startup Moonshot AI has released Kimi K2, a 1-trillion-parameter LLM with open weights and source code.
It uses a Mixture of Experts (MoE) architecture. Instead of activating all parameters at once, it selects just 32 billion that best match the input. This approach yields faster performance, lower computational costs, and higher accuracy.
With a 128K token context window, the model is designed for coding and tool use—it can call APIs, create charts, analyze data, write, debug, and execute code. However, it does not support reasoning mode.
🖥 The model's weights and code are available on GitHub, and it's free to use, even for commercial projects. The only requirement is that if your app has over 100 million users or generates $20M+ per month, you must display the name Kimi K2 in your UI.
Chinese startup Moonshot AI has released Kimi K2, a 1-trillion-parameter LLM with open weights and source code.
It uses a Mixture of Experts (MoE) architecture. Instead of activating all parameters at once, it selects just 32 billion that best match the input. This approach yields faster performance, lower computational costs, and higher accuracy.
With a 128K token context window, the model is designed for coding and tool use—it can call APIs, create charts, analyze data, write, debug, and execute code. However, it does not support reasoning mode.
🖥 The model's weights and code are available on GitHub, and it's free to use, even for commercial projects. The only requirement is that if your app has over 100 million users or generates $20M+ per month, you must display the name Kimi K2 in your UI.
This media is not supported in your browser
VIEW IN TELEGRAM
OpenAI Launches ChatGPT Agent
OpenAI has introduced ChatGPT Agent, an AI assistant designed to handle complex tasks on a virtual computer.
It can browse websites, use the command line, write code, work with spreadsheets, and create slide presentations. At its core, it combines features from the previously introduced Operator tool and Deep Research mode.
The agent is powered by a custom model trained to use tools intelligently and efficiently. The feature is now rolling out to paid users on the web.
OpenAI has introduced ChatGPT Agent, an AI assistant designed to handle complex tasks on a virtual computer.
It can browse websites, use the command line, write code, work with spreadsheets, and create slide presentations. At its core, it combines features from the previously introduced Operator tool and Deep Research mode.
The agent is powered by a custom model trained to use tools intelligently and efficiently. The feature is now rolling out to paid users on the web.
ChatGPT Hits 2.5 Billion Daily Prompts
OpenAI reports that users now send approximately 2.5 billion prompts to ChatGPT daily, with 330 million of those originating from the United States. That's a 150% jump from the 1 billion daily prompts CEO Sam Altman cited eight months ago.
This surge means ChatGPT now handles roughly one-fifth of Google's estimated 13.6–16.4 billion daily searches.
OpenAI reports that users now send approximately 2.5 billion prompts to ChatGPT daily, with 330 million of those originating from the United States. That's a 150% jump from the 1 billion daily prompts CEO Sam Altman cited eight months ago.
This surge means ChatGPT now handles roughly one-fifth of Google's estimated 13.6–16.4 billion daily searches.
ChatGPT Learn
OpenAI has launched ChatGPT Learn
It’s a brand-new built-in learning system:
✅ Learn Python, SQL, ML, writing, productivity
✅ Guided lessons from GPT itself
✅ No signups, no fluff
✅ Free for GPT-4 (ChatGPT Plus) users
🔍 Don’t see the “Learn” tab yet?
It’s rolling out gradually - be patient. It’ll appear in your sidebar soon.
🧠 Get ready. August might just be the biggest AI month yet.
OpenAI has launched ChatGPT Learn
It’s a brand-new built-in learning system:
✅ Learn Python, SQL, ML, writing, productivity
✅ Guided lessons from GPT itself
✅ No signups, no fluff
✅ Free for GPT-4 (ChatGPT Plus) users
🔍 Don’t see the “Learn” tab yet?
It’s rolling out gradually - be patient. It’ll appear in your sidebar soon.
🧠 Get ready. August might just be the biggest AI month yet.
This media is not supported in your browser
VIEW IN TELEGRAM
🎨 ChatGPT just got a creative upgrade: Image Styles are here!
🛠 You can now choose from a variety of visual styles before generating an image—making the creative process more intuitive and controlled.
🔰 This feature is available for both free and Plus users.
📸 You can also upload your own images, and ChatGPT will transform them into your chosen style—while keeping the original content recognizable.
🌟 A big win for designers, creators, and visual storytellers!
🛠 You can now choose from a variety of visual styles before generating an image—making the creative process more intuitive and controlled.
🔰 This feature is available for both free and Plus users.
📸 You can also upload your own images, and ChatGPT will transform them into your chosen style—while keeping the original content recognizable.
🌟 A big win for designers, creators, and visual storytellers!
🧠 GPT‑5 rumored for August release
OpenAI plans to debut GPT‑5 as early as August 2025, possibly as a unified multimodal reasoning system incorporating the o‑series models.
McNeece Web Design+15Reuters+15Tom's Guide+15
🔬 Rumors swirl around GPT‑5’s capabilities
Early testers say GPT‑5 excels in reasoning and software engineering—rivaling Claude Sonnet 4 and perhaps using a dynamic model routing architecture
OpenAI plans to debut GPT‑5 as early as August 2025, possibly as a unified multimodal reasoning system incorporating the o‑series models.
McNeece Web Design+15Reuters+15Tom's Guide+15
🔬 Rumors swirl around GPT‑5’s capabilities
Early testers say GPT‑5 excels in reasoning and software engineering—rivaling Claude Sonnet 4 and perhaps using a dynamic model routing architecture
🧠 New open-weight LLMs: gpt-oss-120B & 20B
OpenAI has just launched those two new models from the gpt-oss project.
I have already tested them!
They’re positioned as open, local alternatives to GPT-4-class models - and yes:
This of course means they are completely free ✅
Here’s the breakdown:
🔹 gpt-oss-20B – smaller, easier to run locally (with a beefy GPU). Decent for coding, Q&A, and experimentation.
🔹 gpt-oss-120B – huge model aiming for GPT-4-like reasoning. Needs serious hardware (128GB+ VRAM), but shows promising results for a fully offline model.
⚠️ Let’s be honest:
They’re not better than GPT-4o - slower, less nuanced, and less aligned out of the box. But for a local, no-API setup, they’re a huge step forward for open source LLMs.
Some facts:
• First open‑weight release since GPT‑2- they’re fully downloadable under Apache 2.0.
• 20B runs on a 16 GB GPU, 120B needs something massive (80 GB+).
• They actually think step by step. Not as sharp as GPT‑4, but surprisingly solid - 120B competes with o3/o4‑mini.
If you want to know more: https://openai.com/index/introducing-gpt-oss/
OpenAI has just launched those two new models from the gpt-oss project.
I have already tested them!
They’re positioned as open, local alternatives to GPT-4-class models - and yes:
they actually run on your own machine (with enough hardware).
This of course means they are completely free ✅
Here’s the breakdown:
🔹 gpt-oss-20B – smaller, easier to run locally (with a beefy GPU). Decent for coding, Q&A, and experimentation.
🔹 gpt-oss-120B – huge model aiming for GPT-4-like reasoning. Needs serious hardware (128GB+ VRAM), but shows promising results for a fully offline model.
⚠️ Let’s be honest:
They’re not better than GPT-4o - slower, less nuanced, and less aligned out of the box. But for a local, no-API setup, they’re a huge step forward for open source LLMs.
Some facts:
• First open‑weight release since GPT‑2- they’re fully downloadable under Apache 2.0.
• 20B runs on a 16 GB GPU, 120B needs something massive (80 GB+).
• They actually think step by step. Not as sharp as GPT‑4, but surprisingly solid - 120B competes with o3/o4‑mini.
If you want to know more: https://openai.com/index/introducing-gpt-oss/
Openai
Introducing gpt-oss
gpt-oss-120b and gpt-oss-20b push the frontier of open-weight reasoning models
🚀 ChatGPT‑5 is live!
🗓️ Released today – August 7, 2025
🧠 What’s new?
• Smarter and faster than GPT‑4
• Handles text, images, audio & video
• Massive context – up to 256k tokens
• Improved coding, writing & health support
• Safer, more accurate responses
📱 Who gets access?
• Free users – limited GPT‑5
• Plus & Pro users – full access (Pro gets GPT‑5 Pro)
• API & Teams – available now
• Enterprise & Education – next week
🔧 Models:
gpt-5, gpt-5-mini, gpt-5-nano
It’s the biggest upgrade since GPT‑4.
🗓️ Released today – August 7, 2025
🧠 What’s new?
• Smarter and faster than GPT‑4
• Handles text, images, audio & video
• Massive context – up to 256k tokens
• Improved coding, writing & health support
• Safer, more accurate responses
📱 Who gets access?
• Free users – limited GPT‑5
• Plus & Pro users – full access (Pro gets GPT‑5 Pro)
• API & Teams – available now
• Enterprise & Education – next week
🔧 Models:
gpt-5, gpt-5-mini, gpt-5-nano
It’s the biggest upgrade since GPT‑4.
Open AI future releases
The next few weeks might change everything (again):
🔥 What’s coming soon?
– GPT-5
– GPT-5 Pro (Zenith)
– GPT-5 High (Summit)
– GPT-5 Mini & Nano (Starfish)
– Lobster (Open-source model)
– Sora 2 (Video generation upgrade)
– Aura (AI-first browser)
– Study Together (collaborative learning)
The next few weeks might change everything (again):
🔥 What’s coming soon?
– GPT-5
– GPT-5 Pro (Zenith)
– GPT-5 High (Summit)
– GPT-5 Mini & Nano (Starfish)
– Lobster (Open-source model)
– Sora 2 (Video generation upgrade)
– Aura (AI-first browser)
– Study Together (collaborative learning)
❤1