Not only you see the final rank and results, but also a comprehensive report with many insights
wdwind@wdwind
Great work! This is my favorite niche bench
English
wdwind
103 posts
Great work! This is my favorite niche bench
FoodTruck Bench@foodtruckbench
DeepSeek V4 Pro just matched GPT-5.2 on FoodTruck Bench, our agentic benchmark — 10 weeks later, ~8× cheaper. First Chinese model in our frontier tier. The China–US gap that used to feel like a year is now ~10 weeks. foodtruckbench.com/blog/deepseek-…
English
More worried now
this wasn't a careless pre-release github push by an intern, core developers have hyped the paper publicly, and now have scrubbed all traces of it. Vibes are off.
Optimistically, they were just notified of prior art and are rewriting the paper. If not?…
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex
Actually worried about what's going on with DeepSeek. They often allow themselves to horse around with the Web/app, it's a (decently made) free sandbox. But their API is a serious business. Hot-swapping models back and forth, rolling out unannounced features… not like them.
English
gemini 3.1 Pro is pretty good
I suspect flash too
English
OK damn, this is getting somewhere
Which models outside the frontier can do it? I haven't tried.
OedoSoldier@OedoSoldier
English
wdwind retweetledi
🚀 DeepSeek-V4 Preview is officially live & open-sourced! Welcome to the era of cost-effective 1M context length.
🔹 DeepSeek-V4-Pro: 1.6T total / 49B active params. Performance rivaling the world's top closed-source models.
🔹 DeepSeek-V4-Flash: 284B total / 13B active params. Your fast, efficient, and economical choice.
Try it now at chat.deepseek.com via Expert Mode / Instant Mode. API is updated & available today!
📄 Tech Report: huggingface.co/deepseek-ai/De…
🤗 Open Weights: huggingface.co/collections/de…
1/n
English
My first few hours in China, first impressions
At the invitation of XPeng, I'm visiting China and starting tomorrow, I'll be attending several events to see the future of robotics and the automotive industry, and of course, learn quite a bit about AI.
My plane landed around 4 PM; jet lag is definitely taking its toll. My first impression from the plane: construction cranes everywhere. China is really building on a massive scale.
Second impression in a Chinese Uber: electric vehicles probably make up 90% of the cars here. Germany, as a "car nation," can't compete with that. EVs are the standard, not the exception.
Third impression: Digitization everywhere. Everything is digital. As someone from Germany who still has to do a lot by mail, it was practically a culture shock. China is miles ahead when it comes to digitization. Everything runs via QR codes and scans.
Fourth impression: At least in the big cities, the infrastructure is modern. Yes, there's a lot of traffic noise and congestion. But the roads are well-developed, the network is functioning, and it's modernly designed.
Tomorrow I'll see much more. It remains exciting!
English
What I’m curious about is this: could DeepSeek V4 take the #1 spot in coding among open-source models, even if it’s not quite at Claude’s level?
I still want to believe The Information’s report that V4 showed strong results in coding.
Xiuyu Li@sheriyuo
From an interview with a mysterious source: DeepSeek V4 is indeed scheduled to be released next week. The delayed release of DeepSeek V4 has nothing to do with Huawei Ascend; it is purely because the results were not satisfactory. The recent financing is also completely unrelated to this matter. With Kimi, Minimax, and Z.ai all going public, Wenfeng Liang has begun to feel anxious. Opinions are my own.
English
wdwind retweetledi
Netanyahu is everything they told us Hilter was.
English
