OpenAI has had an absolutely massive week.
• Best image generation model in the world
• Smartest model in the world
• Custom agents (Enterprise/Business)
• Best image generation model in the world
• Smartest model in the world
• Custom agents (Enterprise/Business)
👾48
No event. No hype. Just two models on Hugging Face
What they shipped:
🔥 V4-Pro — 1.6T params, 49B active
⚡ V4-Flash — 284B params, 13B active
📏 1M context. Both. By default. Free.
Benchmarks:
✅ Open SOTA on agent coding
✅ Leads in math and STEM
✅ Only trails Gemini 3.1 Pro on general knowledge
Old models die July 24, 2026. 3 months to migrate.
📄 https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/DeepSeek_V4.pdf
🤗 https://huggingface.co/collections/deepseek-ai/deepseek-v4
#DeepSeek
Please open Telegram to view this post
VIEW IN TELEGRAM
Please open Telegram to view this post
VIEW IN TELEGRAM
👾44
🚨 April was absolutely insane for AI releases...
• Claude Opus 4.7 (Anthropic)
• GPT-5.5 (OpenAI)
• DeepSeek V4 (DeepSeek)
• Xiaomi Mimi V2.5
• Qwen3.6-Plus (Alibaba)
• GLM-5.1 (Zhipu)
• Muse Spark (Meta)
• Qwen3.6-35B-A3B (Alibaba)
• Grok 4.3 Beta (xAI)
• Qwen3.6-Max (Alibaba)
• Kimi K2.6 (Moonshot)
• Qwen3.6-27B (Alibaba)
• Claude Opus 4.7 (Anthropic)
• GPT-5.5 (OpenAI)
• DeepSeek V4 (DeepSeek)
• Xiaomi Mimi V2.5
• Qwen3.6-Plus (Alibaba)
• GLM-5.1 (Zhipu)
• Muse Spark (Meta)
• Qwen3.6-35B-A3B (Alibaba)
• Grok 4.3 Beta (xAI)
• Qwen3.6-Max (Alibaba)
• Kimi K2.6 (Moonshot)
• Qwen3.6-27B (Alibaba)
👾47
Open Ai
Photo
So you need 2xH200 gpu to run DeepSeek V4 Flash.
Not Pro. Flash.
Or 4x 6000
Or 8x 5090
Or Mac M3 Ultra
Or 2 Frameworks
Not Pro. Flash.
Or 4x 6000
Or 8x 5090
Or Mac M3 Ultra
Or 2 Frameworks
👾4
MoonShotAi
Moonshot AI is rolling out Kimi K2.6 on Kimi Chat and APIs. All models got upgraded. - Kimi K2.6 Instant - Kimi K2.6 Thinking - Kimi K2.6 Agent - Kimi K2.6 Agent Swarm Did you get it already? 👀
MOONSHOT AI LAUNCHED Kimi K2.6 Agent Swarm 👋
Highlights：
🔹 300 parallel sub-agents × 4,000 steps per run (up from 100 / 1,500 in K2.5).
🔹 Outputs are real files, not chat - one run delivers 100+ files, 100,000-word literature reviews, or 20,000-row datasets.
🔹Heterogeneous skills - search, analysis, coding, long-form writing, and visual generation all running in parallel
🔗Try it at: kimi.com/agent-swarm
Highlights：
🔹 300 parallel sub-agents × 4,000 steps per run (up from 100 / 1,500 in K2.5).
🔹 Outputs are real files, not chat - one run delivers 100+ files, 100,000-word literature reviews, or 20,000-row datasets.
🔹Heterogeneous skills - search, analysis, coding, long-form writing, and visual generation all running in parallel
🔗Try it at: kimi.com/agent-swarm
👾44
Open Ai
Photo
It's interesting how all these new AI model releases are now comparing benchmarks with Opus 4.6 and GPT 5.4
This is how far ahead Opus 4.7 and GPT 5.5 are from the competition today.
This is how far ahead Opus 4.7 and GPT 5.5 are from the competition today.
👾38
JUST NOW: Google announces it will invest up to $40 billion in Anthropic, its largest single AI investment ever.
Important details:
Important details:
1: Google initially commits to $10 billion in cash upfront as part of the deal
2: Google can invest up to $30 billion more in Anthropic if performance targets are hit, bringing the total potential commitment to $40 billion
3: Google will supply Anthropic with at least 5 gigawatts of computing power to support AI development
4: SoftBank is separately planning to invest up to $40 billion in Anthropic as well, meaning Anthropic could receive $80 billion in combined commitments from two investors alone
The AI infrastructure race is accelerating.
Two of the world’s largest investors are now betting $80 billion combined on a single AI company.
👾46❤1
JUST IN: State Department launches global campaign targeting DeepSeek over alleged theft from American AI labs.
👾39
JUST IN: Apple’s base M4 Mac Mini is sold out in the U.S., with some other models facing month-long waits.
👾31
everyone is sleeping on NotebookLM...
this thing is literally the best option for:
- building a custom AI trained on whatever topic you need (content creation, copywriting, automations...)
- learning complex topics, studying at any level
- understanding RAG, working with large files from any source
plus it's powered by a model that was clearly designed for these use cases
this thing is literally the best option for:
- building a custom AI trained on whatever topic you need (content creation, copywriting, automations...)
- learning complex topics, studying at any level
- understanding RAG, working with large files from any source
plus it's powered by a model that was clearly designed for these use cases
👾48
According to Breaking Defense, the Department of Defense is using GenAI.mil, a Google Gemini-powered platform that lets Pentagon staff build custom AI agents for routine work on unclassified systems.
What these AI agents are doing:
• Drafting after-action reports
• Writing staff estimates for operations
• Automating paperwork and workflows
• Handling data-heavy admin tasks
• Assisting non-technical personnel with custom tools
This is one of the clearest signs yet that governments are moving from testing AI to deploying AI at scale. Instead of waiting for engineers, everyday Pentagon workers can now “vibe code” their own assistants using natural language.
That means millions of military and civilian employees could eventually build internal software themselves.
Source.
@CoreAti
Please open Telegram to view this post
VIEW IN TELEGRAM
👾44❤3