LLM
-
New Models
Moonshot ships Kimi K3.1 Pro and says cyber training data was stripped before the open-weight drop
Beijing-based Moonshot AI has launched Kimi K3.1 Pro, a post-trained upgrade of its nearly 3-trillion-parameter Kimi K3 with a 1-million-token context window, and says it removed cyber-related training and RL data so the…
-
New Models
Inception launches Mercury 3, a diffusion language model it says matches GPT-6 Luna and Claude Haiku 5.5 at more than 1,000 tokens per second
Inception released Mercury 3 on October 19, 2026, its largest diffusion-based language model, with a 1M-token context window and a claimed speed of 1,068 tokens per second. The company says it matches or beats OpenAI's a…
-
New Models
MemTensor open-sources MemT-1.0, two models that store conversation history in a fixed-size internal memory instead of a growing context
MemTensor has released MemT-1.0-Nano and MemT-1.0-Flash, open-weight models built on Qwen and DeepSeek backbones that write past turns into a learned memory state rather than rereading them. The company reports large sto…
-
New Models
MiniMax launches M3.1 Flash and reveals the anonymous Space Bunny model was an early version of it
Chinese AI company MiniMax has released M3.1 Flash, a low-cost model it says is much stronger at agentic work, coding, computer use and cybersecurity, runs nearly three times faster and cuts agentic task costs by 70%. Th…
-
New Models
Liquid AI releases LFM2.5-24B-A2B, an open-weight reasoning model built for fast on-device agents
Liquid AI has published LFM2.5-24B-A2B on Hugging Face, a 24-billion-parameter mixture-of-experts reasoning model that activates about 2 billion parameters per token and is trained to call tools inside agent harnesses. I…
-
New Models
Google makes Gemini 4 Argon generally available, undercutting rival frontier models on price
Google released its frontier model Gemini 4 Argon to developers, enterprises and paying Gemini subscribers on October 13, 2026, after a two-week preview limited to cyber defenders. Launch pricing of $2 and $10 per millio…
-
Research
Recuris improves long-horizon agents by evolving memory, not model weights
Recuris keeps the base LLM frozen while evolving working and experiential memory, with reported gains across long-horizon agent benchmarks.
-
Industry
Chinese AI models take OpenRouter’s top three spots in real-world usage
DeepSeek, Tencent and Xiaomi lead OpenRouter’s weekly token rankings, showing a major shift in real-world developer usage across coding and agent workloads.