Gemma 4 26B A4B IT
PulseAugur coverage of Gemma 4 26B A4B IT — every cluster mentioning Gemma 4 26B A4B IT across labs, papers, and developer communities, ranked by signal.
3 day(s) with sentiment data
-
Liquid AI's LFM2.5-24B-A2B model offers fast inference but high RAM needs
Liquid AI has released its LFM2.5-24B-A2B model, a 24 billion parameter mixture-of-experts model that only utilizes 2.45 billion parameters per token. This architecture allows for faster decoding speeds, with Liquid rep…
-
Liquid AI releases LFM2.5-24B-A2B for fast on-device agents
Liquid AI has released LFM2.5-24B-A2B, a 24 billion parameter Mixture-of-Experts (MoE) model designed for on-device agents. This model activates approximately 2 billion parameters per token and is optimized for multi-st…
-
LiquidAI releases LFM2.5-24B-A2B for efficient on-device AI agents
LiquidAI has released LFM2.5-24B-A2B, a new reasoning-focused language model designed for efficient on-device agent deployment. Despite having fewer parameters than competitors like Qwen3.6-35B-A3B, it demonstrates comp…
-
autotrust releases Gemma-4 based decision model in bf16 and NVFP4 quantized versions
The autotrust/GEV-26B-Decide model, based on Gemma-4, has been released in two versions: a standard bf16 model and a quantized NVFP4 version optimized for vLLM. The NVFP4 version significantly reduces memory requirement…
-
New benchmark shows strong link between Base64 generation and AI intelligence
A new benchmark called Encode Bench has revealed a strong correlation between a language model's ability to generate Base64 encoded responses and its intelligence scores. The benchmark, which tests models on tasks requi…
-
AI model 'uncensoring' has unintended side effects, study finds
A new study published on arXiv investigates the unintended consequences of removing refusal mechanisms from open-weight AI models, a common practice for creating "uncensored" versions. Researchers found that ablating re…
-
llama.cpp adds Q8_0 quantization support with ZenDNN backend, boosting performance
A pull request to the llama.cpp project introduces support for Q8_0 quantization within the ggml-zendnn backend. Benchmarks demonstrate significant performance gains, with ZenDNN_Q8_0 achieving up to a 193% speedup over…
-
Fine-tuned Qwen3-VL-8B-Instruct outperforms Claude Opus 4.7, GPT-5.5 on PiSAR benchmark
A new research paper introduces the PiSAR benchmark for evaluating screen-conditioned action prediction. The study found that a fine-tuned Qwen3-VL-8B-Instruct model significantly outperformed frontier zero-shot models …
-
Fine-tuned Qwen3-VL model surpasses GPT-5.5 and Claude Opus on new benchmark
A new benchmark, PiSAR, has been developed to evaluate screen-conditioned action prediction in AI models. The benchmark revealed that a fine-tuned Qwen3-VL-8B-Instruct model significantly outperformed frontier zero-shot…
-
RePoT enhances LLM planning by enabling checkpoint recovery
Researchers have introduced RePoT, a method to improve the reliability of Program-of-Thought (PoT) in large language models. RePoT addresses the issue where a single invalid step in a generated plan can invalidate the e…