AI Briefing#open-source-model
The latest AI and developer news about #open-source-model, with the original source and a short summary.
The latest AI and developer news about #open-source-model, with the original source and a short summary.
The open-source model uses a Gemma 4 backbone and Whisper encoder to handle speech, reasoning, and tool calls, outperforming other open-source systems on voice benchmarks.
The model features a hybrid-linear MoE architecture supporting 256K token contexts and is available on Hugging Face and ModelScope.
Kolibri-1 is a 78B-parameter Mixture-of-Experts model with 3.46B active parameters, optimized for German and English reasoning.
Runs on laptops with 2.45B active parameters, outperforming Qwen3.6 in tool-calling performance
The model features 25B active parameters per token and demonstrates strong performance on agentic benchmarks like Terminal-Bench 4.0 and SWE-bench Pro.
The model scores 59.3 on the internal PI Knowledge Work benchmark at $0.13 per task, undercutting Claude Fable 5 and GPT-5.6 Sol while maintaining competitive quality.
The 5B-parameter model achieves state-of-the-art results on OmniDocBench v1.6 with a score of 96.96 and supports seven distinct parsing families via a unified JSON schema.
It reduces latency by returning probability distributions without generation, achieving calibration metrics of KL 0.204 and Brier 0.097.
Released five models and a technical report under the Apache 2.0 license, covering text, speech, and long-form translation.
The new Cyber-OSINT model features 26B total parameters with 4B active, a 262K context window, and is trained on 6,500 OSINT/CTI instructions.
The model uses a VeriLoop-Governed Recurrence (VGR) rule that prevents committing state changes that violate any protected obligation, even if total error counts decrease.
Released six open-weight models and a Studio for testing that return the probability of the optimal choice among candidates instead of generating text.
GGUF quantization files for the K2-Horizon model family have been released, but a forked version of llama.cpp is currently required.
inclusionAI has released Realtime-Venus, a 9B-scale multimodal dialogue model capable of processing video and audio in real time.
UkisAI released the Swift model, which reduces unnecessary inference tokens in Qwen 3.8 27B by 58% and improves speed by 1.95x.
InternLM has released Intern-S2-397B, a 397B-parameter multimodal model designed for scientific problem-solving and long-horizon agent tasks.
OpenUI has released OUI-1, a model fine-tuned from DiffusionGemma to generate UIs using OpenUI-Lang.
The new small LLM Spark-X2.5, featuring a proprietary architecture, has been released, with the 4B model demonstrating performance comparable to Qwen 3.5 9B in benchmarks.
Alibaba's AI model downloads reached 3 billion, surpassing Meta and Google.
The next-generation open model Qwen3.8-27B, with enhanced coding and vision capabilities, has been released.
Qwen has released the Qwen3.8 open model with a total size of 2.4T and 95B active parameters.
NVIDIA is partnering with the local AI community to expand the ecosystem of open models and agents.
Wan-Animate-2 supports video-based character animation and text-based viewpoint control.
InclusionAI released a 124B MoE model with 5.1B active parameters.