🚨 Gemini 4 Argon is now showing up in Google Cloud Once models appear here, launch is normally the same day unless something changes

What 65K INR gets you in 2026 vs in 2024:

It's pretty funny that it seems the one thing Mistral Large 4 is actually leading on is a benchmark about regulation Another EU banger

Day 1: We just got around a 50% speed boost for GPT-6 Astra and GPT-6.1 Sol, pushing them to roughly 50 tokens/s. That’s great. We’re now about halfway to Opus 5.5 speed. I think we deserve a reset.

The real story is this one

All this confirms is that Claude’s API is overpriced lol. 😆

Don't be misled by API pricing!!! Opus 5.5's $1,178 API cost drops to $589 at Sol's prices, and to $213 if it were as token-efficient as Sol 6.1 On the $20 sub, you get roughly the same usage from both models. Opus only looks pricier: higher price per token, more tokens used.

How it feels to ask the Chinese ai models for help on something Claude won’t agree to

As suspected, Anthropic's subscriptions offer over 5X the value of OpenAI's (vs the equivalent API pricing) when using mid-tier models OpenAI offers very slightly more usage of Astra vs Anthropic's Fable 5.5 offering in Pro & Max 20x plans respectively Choose wisely!

Deepseek-V4.1-Flash-2-Sparks is out! It took nearly a month to get this done right - 2000 tok/s prefill - 29 tok/s decode prose - 41 tok/s decode code - 2 million kV cache pool - 262k context window - 8 concurrency - vision - 92% top token agreement - 0.07 KLD - load with pi

anthropic is coming for everyone fable 5.1 is currently routing to fable 5.5 and been testing it for the last 2 days it's gonna be new sota model and from all the testing so far, model is: > faster than current model > super smart > less prompting gives better results, especially for tasks that need intelligence or creativity > still expensive tho at this point openai better use their new base model for astra 6.1, that's the only way they can get back on top anyway super excited for fable to officially release, gonna be super soon iykyk

This isn't bad at all. Dimensity 9600 Pro 4200+ Single core 12700+ Multi core Device: Oppo Find X10 Pro Max

Dimensity 9600 Pro CPU & GPU Performance: Geekbench 6 CPU: 4291 Single core 12807 Multi core The Overall Power Consumption is better than Snapdragon 8 Elite Gen 5, A19 Pro & it's predecessor Dimensity 9500. Geekbench 7 CPU: Hits almost 12000 Multi-core Score & the Power Consumption is better than the 3nm Chips by a good Margin. Also do note, in Geekbench 7 the Snapdragon 8 Elite Gen 5 is better than the A19 Pro in Multi-Core Power Consumption in Mid-High Load Tasks & the gap is more than Geekbench 6. 3D Mark Steel Nomad Light: Dimensity 9600 Pro scores a healthy 4475 Points, Quite a lead over it's Predecessor Dimensity 9500 & other 3nm Chips namely the Snapdragon 8 Elite Gen 5 & Apple A19 Pro.

Snapdragon 8EEG6 powered devices are reportedly facing crashes, black screens and random reboots, with the issue likely becoming more evident under heavy loads. All 8EG6 models including the Honor Magic 9 Pro Max, Xiaomi 18 Pro Series and iQOO 16 have reportedly experiencing similar issues. Of course, not all users are experiencing these problems but a significant number of users have reported encountering them. According to some reports, there may be a flaw in TSMC’s latest 2nm process. The issue is reportedly not limited to Snapdragon powered devices, as the iPhone 18 Pro Series is also facing similar problems. Thankfully, there have been no major issues reported with Dimensity 9600 Pro powered devices. The root cause is still unclear. Hopefully, it can be fixed through OTA updates but if it turns out to be a manufacturing defect, it could become a major headache for the brands. For now, I wouldn’t recommend buying 2nm powered devices blindly until the issue is resolved.

I haven’t touched Claude Code or Codex CLI in a while. The terminal era is over imo. It's the wrong interface for coding agents. Tabs are ephemeral, but context is persistent, and managing 30 tabs is pure cognitive overhead. I don’t really need an IDE like Cursor either. I rarely navigate the whole codebase anymore. The new primitive is the agent, not the file. (Codex desktop app is the best agentic UI for now. But we’re still early.)

🚨 Kirin 9050 Pro is an Engineering MARVEL! You've never seen a Chip like this with Double Die fit on top of Each other. Let me deep dive about this new Double Die & Logic Folding Design. So, Kirin 9050 Pro features double Die Layout. Each takes an Area of 120.6mm² & Combined at 241.3mm². For Reference Kirin 9030 Pro has a Chip Area of 141.7mm². It's about a 70% more Area if the Double Die is Considered. But how is the Double Die Structurally Possible?? Huawei & Hisilicon made this Possible. The Dies are connected through Vertical Copper to Copper Hybrid Bonding. Huawei says the First-generation implementation uses approximately 1.5 µm hybrid-bonding Pitch and around 50 Million vertical Connections between the Layers. It uses TSVs (through Silicon Vias) to connect the two Dies for Electricity/signal flow & a Vertical Connection. For Kirin 9050 Pro, Hisilicon uses 80,000 of these TSVs to connect the Chip to the Substrate & these TSVs takes about 8% of usable Area. The Primary die (the first Half) contents Execution Units such as Integers, Floating Point Units & Load/store Units & the Secondary contents necessary Caches for the CPU. These are linked directly by Extremely dense Metal Layers & Copper-to-Copper Hybrid Bonds. 🔶 Main Die — This contains the more computationally intensive portions: CPU GPU NPU DSP ISP Memory Controller Modem SLC Controller L/S (Load/Store) INT (integer/interrupt-related logic) FPU 🔶 Secondary Die — The left die is primarily devoted to cache and I/O, including: L1 / L2 / L3 cache SLC — 12 MB ISP DSP NPU LPDDR PHY USB 3.2 RF-I/F Various PLL / I/O blocks This 3D Folding Design helps shortening the Distance for Electric Flow that reduces the Latency & Ultimately yields Lower Power Consumption & Lower Voltage at the Same relative Frequency. Bottom line, The Kirin 9050 Pro's Logic-folding / Two-die design separates the low-power, Area-Heavy cache & I/O logic from the High-power Compute logic, allowing Huawei to pack more functionality into the Package while concentrating the thermally demanding CPU/GPU/NPU blocks on the Main die.

Hmm 🤔 Isn't it too Nasty??

The Huawei Mate 90 RS Ultimate Design looks Undeniably Attractive!

I keep hearing that OAI's first generation chip, Jalapeño, is having yield issues.... +There are also rumors that they're pulling in the second generation because of it.

This is the Absolute Power of Snapdragon 8 Elite Extreme Gen 6 ran on Redmagic 12 Pro+. Geekbench 7: