Switchboard Tool Automates LiteLLM Proxy Model Swapping
Switchboard is a community-built tool that keeps a stable alias for each model family in a LiteLLM proxy, so swapping between local backends like vLLM, llama.cp…
Switchboard is a community-built tool that keeps a stable alias for each model family in a LiteLLM proxy, so swapping between local backends like vLLM, llama.cp…
Gemini 4 Argon achieves a new record on the WANDR benchmark and is now available to Perplexity Pro and Max subscribers.
One week after launch, StepFun's Step 5 Preview ranks #2 on OpenRouter's weekly leaderboard with 19.5T tokens used, and has held #2 on the daily leaderboard sin…
Ollama is a day 0 partner for Reflection AI's first open-weight model Beam, offering free limited-time access with a 1M token context window, US hosting, and fu…
Reflection AI released Beam across platforms with open weights under Apache 2.0, including weights, tokenizer, reference implementation, BF16/FP8/NVFP4 variants…
PrismML released Bonsai 2 in 2B, 4B, and 9B sizes built on Qwen3.5, offering ternary weights at 0.43GB, 0.95GB, and 2.0GB for local assistants with improved rea…
Krea 3, Krea AI's new image editing model, is available on OpenRouter. It accepts 1-4 reference images plus a prompt, returns native 2K edited images, and compl…
Weekly token usage ranking of Chinese LLM brands for Oct 9–15: DeepSeek leads with 1.03T tokens, Xiaomi rises to No.2 with +49.9% week-over-week, Zhipu +44.3%,…
Tencent Hy launched Hy Memory 2.0, an open-source memory model, framework, and MemHub tool. The framework manages chat, work, and code memories, ranks #1 open-s…
Tencent open-sourced a full memory stack — the Hy-Memory-30B-A3B model, the Hy Memory engine, and the MemHub app — claiming it leads MemoryArena by 10.2 points…