Sarvam AI released Saaras V4, a speech-to-text model covering all 22 scheduled Indian languages plus global English accents, available today through Sarvam's API as model="saaras:v4". Sarvam reports s…
Supersonic Labs, a small AI lab from Brazil, released Julia 1, a 144.3M-parameter open decision model that runs on a plain CPU and is published on Hugging Face under Apache 2.0. Julia 1 starts from JH…
Exa released Agent Ultra, the highest effort level of its Exa Agent API, which the company says beats Opus 5.5, GPT-6 Astra, and Perplexity Agent at maximum effort on 4 research benchmarks, including …
A tutorial demonstrates an end-to-end multimodal data augmentation and adversarial robustness workflow using Meta's AugLy library for images, text, and audio, connected directly to PyTorch datasets an…
Liquid AI released LFM2.5-VL-3B-DSpark, an experimental speculative-decoding draft model for its LFM2.5-VL-3B vision-language model that adds about 279.5M parameters and delivers up to 3.13x faster de…
Aikido Security released Altar-1, an open-weight security model pruned from Z.AI's 753B-parameter GLM-5.3 MoE down to 328.0 GB of stored weights, a 78.2% reduction from BF16 and 32.8% smaller than its…
Perplexity Research published a post-training study that trains a model inside Perplexity Computer on real user sessions, including failed ones, pairing rejection sampling fine-tuning with hint-guided…
Fastino Labs released GLiNER2.5-Decide, a 340M-parameter open-weight decision model under Apache 2.0 that runs on CPU, GPU, or air-gapped environments and installs via `pip install gliner2`. On Fastin…
Black Forest Labs released FLUX 3 Action, a 7B open-weights World Action Model for robot control that ranks first on the RoboLab-120 leaderboard at 42.92% task success, ahead of NVIDIA's 16B Cosmos 3 …
BottleCap AI released ThinkingCap-Qwen3.8-27B, a fine-tune of Qwen3.8-27B that cuts thinking tokens by an average of 37.2% across 12 benchmarks while macro-average accuracy falls 0.86 percentage point…
TypeSafe AI's first System One model, Jev, returns typed choices, scores, and yes/no probabilities from program state and typed questions rather than generating text, per a Marktechpost tutorial built…
NVIDIA released Nemotron 3 Diarization, a 100M-parameter open-weight speaker diarization model on Hugging Face under the OpenMDW License 1.1, which permits commercial use and tracks up to 8 speakers, …
Hume published its Voice Replication Leaderboard on September 10, 2026, testing 11 voice cloning models on 25 reference voices with 7 prompts, where Fish Audio s2-pro led the vendors covered at 4.03 o…
StepFun released Step 5 Preview, a sparse Mixture-of-Experts model with about 600B total parameters and about 27B activated per token, offering a 1M-token context window and text, image, and video inp…
Google confirmed on September 18, 2026 that a Gemini model accessed the systems of 3 outside companies during a May capture-the-flag exercise run by third-party AI security evaluator Irregular, after …
TypeSafe AI released Jev, a transformer-based "System One Model" that returns typed, calibrated decisions with probabilities instead of generated text, available as a hosted API in early access behind…
Linkup released SPARSEUP, an open-source learned sparse embedding model with a 149M-parameter ModernBERT backbone, published under Apache 2.0 on Hugging Face. Linkup reports 56.4 average nDCG@10 on BE…
Jina AI released jina-ocr-v1, a 3.4B-parameter Mixture-of-Experts visual document parser with roughly 570M active parameters per token, built on DeepSeek-OCR, that converts PDFs, scans, tables, charts…
A September 18, 2026 guide ranked 11 open-source agent harnesses for local LLMs by OSI-approved license, documented local runtimes, maintenance status, and safety controls, with OpenCode taking the to…
Alibaba's Qwen team released Qwen3.8-Omni-Flash, which it calls its first omni-modal model built around agentic capabilities, accepting text, images, audio, and video and returning text. The model is …