multi-language

Benzi is a compiler-backed AI coding agent that uses deterministic static analysis and code indexing instead of reading raw source code, reporting 78.2% on SWE-bench Verified for under 10 cents per fix while reading far fewer tokens than Claude Code, DeepSeek's harness, and OpenCode.

Datalab's Marker v2 is an open-source document parsing tool that handles PDFs, images, DOCX, and PPTX, outputting clean Markdown and supporting over 90 languages for batch organization, knowledge bases, and AI document processing.

DemoScreen turns product screenshots into finished demo videos with scripts, voiceover, music, and captions, offering multiple formats and languages with a free plan.

Glasshouse v0.1 is a memory benchmark tool designed to address issues with existing AI memory benchmarks by providing multi-language support, varied conversation lengths, and detailed evaluation across multiple axes.

SOUND is a macOS app that provides per-application audio equalization, volume control, and speaker selection from the menu bar, addressing the lack of built-in per-app audio settings.

Maaa is a macOS app that uses the notch to provide gentle reminders for drinking water, eating, and taking breaks, available in 12 languages for macOS 14 and later.

A developer launched a set of 3D radios powered by GPT-6 Astra on YouWare, offering global internet radio stations with features like random play, favorites, and immersive interactions.

LoudKit is an open-source local TTS tool with voice cloning, supporting 10 languages, and available as SDKs for Python, Swift, Go, Rust, and TypeScript, designed to run efficiently on edge devices.

Benzi is a new agentic harness that minimizes source code reading by using deterministic intelligence, achieving better performance on benchmarks like SWE-bench with lower token usage and cost.

Datalab has released Marker v2, an open-source document parsing pipeline that efficiently converts PDFs, images, DOCX, and PPTX files to markdown with support for over 90 languages and high performance on GPUs.

VoiceStudio is a locally running AI voice tool, offering features such as voice cloning, sound design, video dubbing, etc., supporting 646 languages, with no need for network connection or subscription, and all data processed locally.

Liquid AI's LFM2.5-Encoders can identify and strip 40 types of PII across 16 languages in one pass, designed for secure AI data processing within private pipelines.

Cursor open-sources the SDK Bridge, a protobuf-based protocol that lets developers build Cursor agents in any language (e.g., Rust, Go, Java) without depending on the official TypeScript or Python SDKs.

Fish Audio launches S2.1 Pro, a production voice model with 90ms latency, support for 83 languages, voice cloning from short samples, and multi-speaker dialogue, available via API with a free tier for development.

China has open-sourced the OCR model Unlimited-OCR with only 3B parameters, which can parse an entire 100-page PDF in one go, supports local execution, achieves 93% accuracy, and is completely free and open-source.

Hacker Vitto Rivabella publicly announced the successful jailbreak of Fable 5, analyzing in detail the model's multi-layered security mechanisms, including input/output auditing, intent detection, and chain-of-thought defense, and provided methods to bypass them.

RoadmapBench presents a new benchmark of 115 long-horizon coding tasks across 17 repositories and 5 languages, requiring agents to implement multi-target version upgrades. Evaluation of 13 frontier models shows that even the best, Claude-Opus-4.7, resolves only 39.1% of tasks, highlighting that long-horizon software development remains largely unsolved.

Fusion is a programming language that allows developers to write libraries once and transpile them to C, C++, C#, D, Java, JavaScript, Python, Swift, TypeScript, and OpenCL from a single codebase.

NetEase Youdao open-sourced the 1.3B parameter Confucius4-TTS model, supporting zero-shot voice cloning and cross-lingual speech synthesis in 14 languages, fast and with excellent results.

Multi-LCB extends the LiveCodeBench benchmark to evaluate LLMs across twelve programming languages while preserving contamination controls, revealing Python overfitting and language-specific contamination issues.