Gork and Mork, or Navarro and Ron Vara, I guess
Rumors about the H200 keep taking unexpected turns, you can no longer tell what's true.
Artificial Intelligence thread
- Thread starter 9dashline
- Start date
Rumors about the H200 keep taking unexpected turns, you can no longer tell what's true.
so not only DeepSeek, but ByteDance and Alibaba are launching new models around CNY.
and they are all rushing out red envelopes to get people to use during new year period.
I've been using Kimi a lot in the past couple of days. Really good. I actually find it to be the best free access LLM I tried online. Much better than current DeepSeek and Gemini-3 (at least the free version available online).
This is literally the first time you had a Chinese AI model available that's almost as good as the latest American moderns.
I remember when R1 came out, o3 came out right afterward and o3 was just leagues ahead of R1 in the answers. As far as I can see, OpenAI isn't holding things back anymore so GPT 5.2 is the best they got and it's only a little better than Kimi based on the scores.
Doubao is now up to 63 trillion tokens per day. This is up from 50T per day just a month ago. It has > 100m DAU in China.
The Dola app has 10m DAU globally. It looks like 2026 will be a global expansion period for ByteDance AI.
GLM-5 this month
Step 3.5 Flash is our most capable open-source foundation model, engineered to deliver frontier reasoning and agentic capabilities with exceptional efficiency. Built on a sparse Mixture of Experts (MoE) architecture, it selectively activates only 11B of its 196B parameters per token. This "intelligence density" allows it to rival the reasoning depth of top-tier proprietary models, while maintaining the agility required for real-time interaction.
- Deep Reasoning at Speed: While chatbots are built for reading, agents must reason fast. Powered by 3-way Multi-Token Prediction (MTP-3), Step 3.5 Flash achieves a generation throughput of 100–300 tok/s in typical usage (peaking at 350 tok/s for single-stream coding tasks). This allows for complex, multi-step reasoning chains with immediate responsiveness.
- A Robust Engine for Coding & Agents: Step 3.5 Flash is purpose-built for agentic tasks, integrating a scalable RL framework that drives consistent self-improvement. It achieves 74.4% on SWE-bench Verified and 51.0% on Terminal-Bench 2.0, proving its ability to handle sophisticated, long-horizon tasks with unwavering stability.
- Efficient Long Context: The model supports a cost-efficient 256K context window by employing a 3:1 Sliding Window Attention (SWA) ratio—integrating three SWA layers for every one full-attention layer. This hybrid approach ensures consistent performance across massive datasets or long codebases while significantly reducing the computational overhead typical of standard long-context models.
- Accessible Local Deployment: Optimized for accessibility, Step 3.5 Flash brings elite-level intelligence to local environments. It runs securely on high-end consumer hardware (e.g., Mac Studio M4 Max, NVIDIA DGX Spark), ensuring data privacy without sacrificing performance.
The Top Open AI Models Are Chinese. Arcee AI Thinks That’s A Problem.
The American startup is pitching investors on a $1 billion+ valuation to train a model over a trillion parameters, aiming to reclaim the open-weight lead from Chinese labs like Moonshot and DeepSeek.American investors are confronting an uncomfortable reality in artificial intelligence: the most powerful open models in the world are no longer being built in the United States, but in China. Over the past year, a growing chorus of technologists and financiers has warned that the U.S. is quietly surrendering the open-weight AI market to Chinese labs like DeepSeek, Moonshot AI and Z.ai.
By one widely watched measure, the concern is no longer theoretical. The top six best open models are all developed by Chinese firms, according to independent AI benchmarking outfit Artificial Analysis’ . They’ve been steadily gaining traction: Chinese open models’ weekly usage as a share of total AI usage was 1.2% in late 2024 but surged to nearly 30% in December, according to a published by OpenRouter and the venture capital firm Andreessen Horowitz.
"Roughly 20% of AI startups use open source models, and of those companies, I would say roughly 80% are using Chinese open models,” Andreessen Horowitz general partner Martin Casado told Forbes.
interesting cnbc show talking about AI.. just weird when the Guy interviewed claiming CN is somewhat constrained in energy, using electricity per capita as basis...
Dola has surpassed ChatGPT in app downloads this year to become the top downloaded GenAI app in the top 5 overseas market it has launched in - Indonesia, Mexico, Colombia, Brazil, Philippines. Bytedance's Dreamina (overseas version of Jimeng image/video gen app) has also exploded in downloads
Although OpenAI has poached many Meta employees, Bytedance still has much stronger experience and institutional knowledge in mobile User Acquisition.
Interestingly, Dola has been released in the UK as well and is seeing good traction. Bytedance UA strength should not be underestimated
