Dera News

Open Source AI

Llama, Mistral, Qwen, DeepSeek, and the open-source models that are reshaping AI economics. Self-hosting, fine-tuning, and breaking free from vendor lock-in.

12 articles
Last 30 days
(15 total matching)
Google article thumbnail
Google and 3 others
Google

Meet GitHub Spec-Kit: An Open Source Toolkit for Spec-Driven Development with AI Coding Agents

If you have spent time using AI coding agents — GitHub Copilot, Claude Code, Gemini CLI — you have probably run into this situation: you describe what you want, the agent generates a block of code that looks correct, compiles, and then subtly misses the actual intent. This “vibe-coding” approach can work for quick prototypes The post Meet GitHub Spec-Kit: An Open Source Toolkit for Spec-Driven Development with AI Coding Agents appeared first on MarkTechPost.

Notable
Mistral article thumbnail
Mistral logo
Mistral

Mistral’s Model Lets You Vibe Long-Running Code in the Cloud

Mistral AI has launched its new 'Mistral Medium 3.5' model, integrated into a remote coding agent called 'Vibe' and an AI chatbot 'Le Chat.' These tools now feature a 'Work mode' that allows for extended, autonomous coding and various business tasks using natural language, promising significant improvements in development and operational efficiency for SMBs.

Notable
DeepSeek article thumbnail
DeepSeek logo
DeepSeek

Three reasons why DeepSeek’s new model V4 matters

DeepSeek, a Chinese AI company, has launched its new V4 model, promising high performance at a significantly lower cost than models from OpenAI or Anthropic. This release includes two versions, V4-Pro and V4-Flash, both designed to handle complex tasks and coding efficiently. The competitive pricing makes advanced AI capabilities a powerful and affordable option for small and medium-sized businesses.

Notable
NVIDIA article thumbnail
Anthropic and 1 others
NVIDIA

LightSeek Foundation Releases TokenSpeed, an Open-Source LLM Inference Engine Targeting TensorRT-LLM-Level Performance for Agentic Workloads

Inference efficiency has quietly become one of the most consequential bottlenecks in AI deployment. As agentic coding systems such as Claude Code, Codex, and Cursor scale from developer tools to infrastructure powering software development at large, the underlying inference engines serving those requests are under increasing strain. The LightSeek Foundation researchers have released TokenSpeed, an The post LightSeek Foundation Releases TokenSpeed, an Open-Source LLM Inference Engine Targeting T...

Notable
DeepSeek article thumbnail
DeepSeek logo
DeepSeek

PRISM: Prior Rectification and Uncertainty-Aware Structure Modeling for Diffusion-Based Text Image Super-Resolution

DeepSeek has introduced PRISM, an AI framework utilizing diffusion models to restore low-resolution text images with high accuracy. This technology significantly enhances character recognition, especially for previously unreadable images. It addresses long-standing challenges in maintaining text legibility while improving image quality, offering potential advancements for document digitization and OCR.

NVIDIA article thumbnail
NVIDIA and 1 others
NVIDIA

MISA: Mixture of Indexer Sparse Attention for Long-Context LLM Inference

Institution: Peking University | Authors: Ruijie Zhou, Fanxu Meng, Yufei Xu, Tongxuan Liu, Guangming Lu arXiv Links arXiv | PDF AI summary Abstract MISA replaces the dense token-wise indexing in sparse attention with a routed mixture-of-experts approach that reduces computational cost while maintaining performance and handling long contexts effectively. AI-generated summary 摘要 MISA 用路由专家混合方法取代了稀疏注意力中的密集标记索引,该方法降低了计算成本,同时保持了性能并有效处理长上下文。 AI 生成的摘要 Abstract DeepSeek Sparse Attention (DSA) sets the s...

DeepSeek article thumbnail
OpenAI and 2 others
DeepSeek

China’s DeepSeek previews new AI model a year after jolting US rivals

Chinese AI company DeepSeek has released a preview of its V4 model, aiming to compete with major US AI systems. The company claims V4 offers significant improvements, especially in programming capabilities crucial for AI agents. This launch follows DeepSeek's previous model, R1, which garnered attention for its low training cost, and highlights China's advancements in AI despite ongoing scrutiny over hardware and data use.

Moonshot AI article thumbnail
Moonshot AI logo
Moonshot AI

Moonshot AI Releases Kimi K2.6 with Long-Horizon Coding, Agent Swarm Scaling to 300 Sub-Agents and 4,000 Coordinated Steps

Chinese AI lab Moonshot AI has released Kimi K2.6 as open-source, a multi-modal agent model that significantly advances autonomous software development. It supports long-duration code generation, natural language front-end creation, and coordinates up to 300 sub-agents, offering powerful new capabilities for automating complex programming tasks and boosting development efficiency.

DeepSeek article thumbnail
OpenAI and 2 others
DeepSeek

DeepSeek could hit $45B valuation from its first investment round

Chinese AI lab DeepSeek is making waves by developing powerful large language models with significantly less computing power and cost than its U.S. counterparts. This efficiency has positioned the company for a potential $45 billion valuation in its initial funding round, highlighting a shift in the global AI landscape and the increasing competitiveness of non-U.S. innovators.

OpenAI article thumbnail
OpenAI and 3 others
OpenAI

Cline Releases Cline SDK: An Open-Source Agent Runtime Now Powering Its CLI and Kanban, With IDE Extensions Being Migrated

Cline has extracted its internal agent harness into an open-source TypeScript SDK called @cline/sdk, the same runtime now powering its CLI and Kanban, with VS Code and JetBrains extensions being migrated. The SDK is structured as a four-layer stack — @cline/shared, @cline/llms, @cline/agents, and @cline/core — with native support for plugins, subagents, CRON scheduling, checkpointing, and MCP connectors. On Terminal Benchmark 2.0, Cline CLI scored 74.2% on claude-opus-4.7, compared to Anthropic'...