Models & Labs
Inception Labs tests diffusion LLM Mercury 3 speed and accuracy
Sam Witteveenhigh confidence
Why it matters
- →Diffusion LLMs like Mercury 3 offer significantly higher throughput than autoregressive models, potentially lowering inference costs.
- →Independent benchmarks show trade-offs in reasoning depth, helping practitioners choose the right model for speed vs. accuracy.
- →The integration of vision via Mercury Decide expands the utility of diffusion architectures beyond pure text generation.
Inception Labs has released Mercury 3, a diffusion-based large language model capable of generating text at speeds exceeding 1,000 tokens per second. Independent evaluations by AI researcher Sam Witteveen compared the model against industry standards like GPT-6 Luna and Claude Haiku 5.5, finding that while Mercury 3 matches them on general benchmarks, it lags in complex agentic tasks such as tool calling. The company also introduced Mercury Decide, a companion model with vision capabilities, allowing for multimodal inputs in their pipeline. These models are available via the Inception API and OpenRouter, offering a new architectural approach to high-speed text generation.
Read originalThe story around this
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
More from Sam Witteveen
More in Models & Labs
Models & Labsmodels
llama.cpp b12029 adds CUDA 13 and Snapdragon support
This release quietly cements llama.cpp as the universal inference runtime by finally supporting NVIDIA's latest CUDA 13 stack alongside AMD's ROCm 10.0. The inclusion of native builds for Linux arm64 with Snapdragon chips, covering CPU, Adreno GPU, and Hexagon NPU, signals a serious push into mobile edge deployment. While Apple Silicon KleidiAI is disabled in this specific build, the broader hardware coverage means developers no longer need to compile from source for most modern accelerators. It’s less about new features and more about removing friction for anyone running local models on bleeding-edge or diverse hardware.
llama.cpp Releases
© The Verge AIModels & Labsother
