Models & Labs
Google Launches Gemini 4 Carbon for Efficiency
AI Explainedhigh confidence
Why it matters
- →Expands the practical utility of Gemini models for cost-sensitive applications
- →Highlights industry trend toward specialized model variants (carbon, nano, etc.)
- →Increases competition in the efficient inference market
Google has officially introduced Gemini 4 Carbon, a new addition to its Gemini family of models designed to offer improved efficiency and lower costs for inference tasks. The announcement highlights Google's strategy to segment its model offerings based on performance versus resource consumption, allowing developers to choose models that best fit their specific latency and budget requirements. While details on specific benchmark improvements were sparse in initial reports, the launch signals a competitive push against rivals like OpenAI and Anthropic to provide versatile model tiers. The release has sparked discussion within the AI community regarding the trade-offs between raw capability and operational efficiency.
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 AI Explained
© AI ExplainedResearchresearch
New Benchmarks: ExploitGym, BenchCAD, and IntegrityBench
Several new evaluation benchmarks have been released, including ExploitGym for security vulnerabilities and BenchCAD for design robustness.
AI Explained
© AI ExplainedResearchresearch
METR Report: AI R&D Acceleration at Anthropic
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.
