Models & Labs
Google's Gemini 3.5 Pro Model Still Missing
The Rundown AIhigh confidence
Why it matters
- →Google's delay in releasing the 3.5 Pro model highlights potential competitive challenges in the AI market.
- →The new Gemini models focus on efficiency, but still lag behind key competitors in performance.
- →The ongoing development of Gemini 4 suggests Google is aiming for a significant leap in AI capabilities.
Google has introduced three new models to its Gemini lineup, focusing on speed and cybersecurity, but the anticipated 3.5 Pro model remains unreleased. The new models include the 3.6 Flash, which offers efficiency improvements but still falls short compared to rivals like Grok 4.5 and GPT-5.6 Luna. Google's delay in releasing the 3.5 Pro model raises questions about its competitive position in the AI market. Meanwhile, the company is advancing with its Gemini 4 pre-training, described as its most ambitious yet.
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 The Rundown AI
More in Models & Labs
Models & Labsother
llama.cpp b12340 adds CUDA 13 and ROCm 10.0
This release quietly closes the hardware gap for local inference by adding default support for CUDA 13 and ROCm 10.0 across Linux and Windows. Users with newer NVIDIA GPUs or AMD accelerators no longer need to compile from source to get performance, while Apple Silicon builds now ship KleidiAI disabled by default, shifting the focus back to standard ARM optimizations. The inclusion of Snapdragon support for Linux and Android marks a significant step toward on-device AI beyond the x86/ARM desktop ecosystem. It is less about new features and more about ensuring llama.cpp remains the universal runtime for every available accelerator.
llama.cpp Releases
Models & Labscoding
llama.cpp b12342 adds few-row MMQ for speculative decoding
Speculative decoding just got a significant speed boost on NVIDIA GPUs. This release introduces a specialized kernel for 'few-row' matrix multiplications, targeting the small batch sizes typical of verification passes in speculative decoding workflows. By optimizing these specific operations for quantized formats like Q8_0 and IQ4 variants, llama.cpp reduces latency where it matters most during token generation. The change is CUDA-only for now but leverages hardware capabilities from Ampere onwards, making local inference faster without requiring new model architectures or complex configuration changes.
