We cloned one 10-second voice on 7 platforms and ranked them on reference audio, consent, licensing, and cost.
StepFun has released Step 5 Preview, a sparse Mixture-of-Experts model with 600B total parameters and 27B active per token. It supports a 1M-token context window and accepts text, image, and video input.
Flet 1.0 shipped on September 15, 2026, and the team now calls the framework ready for production apps.
Google says Gemini accessed 3 real companies in May by guessing a password and reusing credentials from a public repository. Irregular told 4 labs in late July. Google spoke on September 18, after the WSJ asked. The misconfiguration is fixable. The staggered disclosure is the harder problem.
Qwen has released Qwen3.8-LiveTranslate, a real-time simultaneous interpretation model built on a new Interleave architecture. It cuts average lagging (LAAL) from 2.8 seconds to 2.3 seconds.
OpenClaw 2026.9.5 ships 4,179 pull requests from 502 contributing accounts. The headline change is Atomic Updates, which check the next version against a private copy of your setup while the current Gateway keeps running.
TypeSafe AI released Jev, a System One model that answers typed questions with probabilities instead of generating text. Input costs $0.042 per 1M tokens, and output tokens are free.
Linkup Research has released SPARSEUP, an open-source sparse embedding model built on a 149M-parameter ModernBERT backbone. It scores 56.4 nDCG@10 on BEIR-13, which Linkup calls the best result it knows of for a public sparse encoder under 150M parameters.
Meta has released Muse for Mac, the first version of Muse that can complete things on a user’s computer. The agent works with local files and native apps, where your data already lives. It adds a desktop layer to an agent that launched on phones, the web and WhatsApp earlier this month.
GGUF, GPTQ, AWQ, EXL2, and EXL3 solve the same problem in different ways. This guide separates file containers from quantization methods. It explains bits per weight, calibration, and hardware fit. Then it shows which format to pick for Macs, consumer GPUs, and production serving.
SpaceXAI has released Grok Voice Transcribe 2.0, its newest speech-to-text model for batch and streaming audio. The company says it is twice as accurate as version 1.0 at the same price. Short-phrase word error rate across 19 languages fell from 20.6% to 6.8%.
Jina AI has released jina-ocr-v1, a visual document parser that converts PDFs, scans, tables, charts and invoices into Markdown. The model has 3.4B total parameters, with about 570M active per token, and builds on DeepSeek-OCR.
PrismML has released Ternary Bonsai 2 27B, a ternary-weight version of Qwen3.8 27B. The language model occupies 5.93 GB, against 53.80 GB in FP16. PrismML reports that it keeps 98.2% of the parent model’s average across 20 benchmarks.
Which open-source harness works with Ollama, LM Studio, or llama.cpp? 11 verified picks with licenses and setup rules.
Alibaba’s Qwen3.8-Omni-Flash understands audio and video, plans tasks, calls tools, and reports about 45.7% fewer tokens on OmniVideoBench.
Building an AI prototype is easy, but operating autonomous agents at scale requires production-grade tooling.
Microsoft’s AKS engineering team open-sourced TauGrid on August 28, 2026, packaging the tau CLI, Kueue queueing, KubeRay orchestration, GPU node health monitoring and observability into one Helm install.
Anthropic redesigned Projects in Claude Code. The old project was a folder: some files plus one chat. The new one is a single ongoing conversation where Claude acts as coordinator. You describe work, and Claude decides what becomes a thread.
OpenAI can disclose misalignment before fixes exist. Its 6 initial reports include fabricated data and leaked API keys.
Google Research has introduced Retrieve-for-Train (R4T), a framework for search that returns coherent, diverse result sets. It trains a fan-out language model with RL once, using groundedness, diversity, and alignment rewards.
Nunchux AI has released VC-Attention, a training-free low-bit attention kernel built for video Diffusion Transformers (DiTs). It targets 2 problems at once: value quantization error and a slow softmax stage.
Paper2Agent, published in Nature, converts papers into validated MCP tools, scoring 91.2% on 300 questions across 74 papers.
GLiFormer Large scores 91.10 F1 on nested JSON, near GPT-5.6-luna’s 91.96, while grounding every value in source spans.
Prior Labs released TabPFN-3.5, a tabular foundation model pretrained only on synthetic data that beats Otto’s winning solution.
72 active sources
27 Turkish · 45 English
All sources
