Pure Neo - AI News Overview

August 14, 2025

Google adds automatic memory and temporary chat controls to Gemini

Google has begun rolling out automatic memory in Gemini AI, allowing the assistant to remember details from past interactions by default. The update also introduces a “Temporary Chat” mode that does not store or use conversations for training and expires after 72 hours. The changes aim to balance personalization with stronger privacy controls.

Read the full article >

August 14, 2025

Google tests Magic View in NotebookLM as potential new data visualization feature

Google is trialing a new experimental feature in NotebookLM called Magic View. The tool displays an animated, dynamic canvas when activated, resembling a generative simulation. While the exact function is unconfirmed, early indications suggest it could provide new ways to visualize and interact with notebook content. The development follows recent updates to NotebookLM that expand multimedia support, including video overviews and mind maps.

Read the full article >

August 14, 2025

Google releases Gemma 3 270M, an ultra-efficient open-source AI model for smartphones

Google DeepMind has released Gemma 3 270M, a compact 270 million-parameter model designed for instruction following and text structuring. The open-source model is optimized for low-power hardware, including smartphones, browsers, and single-board computers. Its efficiency enables AI capabilities in privacy-sensitive and resource-constrained environments.

Read the full article >

August 12, 2025

Claude adds memory search for past conversations

Anthropic has introduced a new memory feature in Claude that lets users search and reference past conversations in new chats. The feature launches today for Max, Team, and Enterprise plans, with availability for other plans expected soon. It aims to streamline workflows by eliminating repeated context-sharing and enabling project continuity.

Read the full article >

August 12, 2025

Anthropic expands Claude Sonnet 4 to 1 M token context

Anthropic has increased the context window for Claude Sonnet 4 to 1 million tokens, allowing the model to process entire codebases or dozens of research papers in a single prompt. The upgrade is available in public beta through Anthropic’s API and Amazon Bedrock, with Google Cloud Vertex AI integration planned. The change enables more coherent large-scale reasoning and workflow automation for developers and enterprises.

Read the full article >

August 12, 2025

Google makes Jules, its AI coding agent, generally available

Google has transitioned Jules from Google Labs beta to full public release. Powered by Gemini 2.5 Pro, Jules runs asynchronously in a cloud VM to read, test, improve, and visualize code with minimal developer oversight. The launch adds a free tier alongside paid “Pro” and “Ultra” plans and introduces a critic capability that flags issues before changes are submitted.

Read the full article >

August 8, 2025

OpenAI Publishes GPT-5 Prompting Guide and Releases Prompt Optimizer Tool

OpenAI has added a GPT-5 prompting guide to its public Cookbook and launched a Prompt Optimizer in the Playground. The resources are designed to improve the quality, efficiency, and consistency of prompts for the new model, with specific guidance for agentic tasks, coding workflows, and instruction adherence.

Read the full article >

August 7, 2025

OpenAI launches GPT-5 with unified fast and reasoning modes for API and ChatGPT

OpenAI has released GPT-5, a flagship language model combining high-speed responses with advanced reasoning in a single system. The model is available immediately in the OpenAI API and to ChatGPT Team users, with Enterprise and Education accounts gaining access next week. GPT-5 introduces a 400,000-token context window, enhanced coding performance, expanded developer controls, and improved reliability over previous models.

Read the full article >

August 5, 2025

GPT‑OSS: OpenAI Publishes 20B and 120B Open‑Weight Models for Local Deployment

OpenAI has released gpt‑oss‑120b and gpt‑oss‑20b, its first open‑weight models since GPT‑2. The models match or exceed the performance of proprietary counterparts and mark a rare moment of open source leadership from a U.S.-based AI lab. With support for tool use, chain‑of‑thought reasoning, and smooth MacBook deployment, gpt‑oss is designed for full local control.

Read the full article >

August 5, 2025

Claude Opus 4.1 Raises Coding and Reasoning Performance

Anthropic has released Claude Opus 4.1, a mid-cycle upgrade focused on coding accuracy, reasoning stability, and better multi-file task tracking. It delivers improved benchmark scores and enhanced real-world usability, particularly for enterprise and developer workflows. The update is live across Claude Pro, API, Bedrock, and Vertex AI with no pricing change.

Read the full article >

August 5, 2025

DeepMind Debuts Genie 3 for Real-Time Text-to-3D World Generation

A new generation of world models is here. DeepMind’s Genie 3 turns text prompts into playable 3D environments at 24 fps with persistent memory and interactive elements. While still in research preview, it represents a major step toward AI agents that can learn and act in open-ended virtual worlds.

Read the full article >

July 31, 2025

Alibaba Shrinks Its Coding AI to Run Locally

Qwen3-Coder-Flash is a compact, 30B MoE model capable of local inference on modern MacBooks. It joins Alibaba’s broader Qwen3 ecosystem as a nimble counterpart to the heavyweight 480B hosted version, giving developers a pragmatic hybrid setup for coding workflows.

Read the full article >

July 31, 2025

NVIDIA Releases Llama Nemotron Super v1.5 to Push Open-Source Agent Reasoning

The new 49B model tops open benchmarks with a 128K context window, tool-use capabilities, and single‑GPU efficiency. It's a signal that NVIDIA aims to lead in agent‑focused LLMs that actually run in production.

Read the full article >

July 31, 2025

Krea Releases FLUX.1 Krea Model with Open Weights

The team behind the FLUX image ecosystem just open-sourced FLUX.1 Krea, a distilled model fine-tuned for photorealistic and aesthetic image generation. Developed with Krea, the model is now freely available under a non-commercial license and slots directly into the broader FLUX.1-dev workflow.

Read the full article >

July 31, 2025

Deep Cogito Releases Cogito v2 Models, Challenging Frontier AI on a Budget

Four open-source language models with advanced reasoning and sub-$3.5M training cost aim to rival top-tier AI. Deep Cogito’s approach blends inference-time search with distilled intuition, offering a compelling alternative to closed frontier models.

Read the full article >

July 31, 2025

StepFun open-sources Step3, a 321B parameter VLM optimized for Chinese AI chips

StepFun has released Step3, a massive open-source visual language model with 321 billion parameters and leading benchmark scores. The model debuts with novel attention architectures that reduce inference costs and is optimized to run efficiently on domestic Chinese AI hardware.

Read the full article >

July 30, 2025

Claude Mobile Adds Email and Calendar Actions

Anthropic’s Claude app can now draft and send emails, messages, and calendar invites directly from mobile. This upgrade turns Claude into a more active assistant for personal and professional tasks. The update strengthens its position as a daily utility for users looking to streamline communication on the go.

Read the full article >

July 30, 2025

n8n Launches No-Setup RAG Template with Vector Store and OpenAI

n8n has released a plug-and-play RAG starter template that lets users upload documents and chat with them instantly. The workflow requires no custom setup and includes preconfigured ingestion and query pipelines using OpenAI.

Read the full article >

July 30, 2025

Ollama Adds Desktop App and Performance Upgrades in v0.10.0

The latest Ollama update brings a native desktop app for macOS and Windows, along with CLI and multi-GPU performance enhancements. It's a move aimed at making local LLM deployment more accessible and efficient across devices.

Read the full article >

July 30, 2025

IBM and Google Launch Rival Toolkits for Document Intelligence

Two enterprise AI contenders just released open-source toolkits aimed at extracting structured insights from unstructured documents. IBM’s Docling and Google’s LangExtract target different parts of the same problem—streamlining how organizations turn documents into usable data.

Read the full article >

Signal vs. Noise

The Pure Neo AI Timeline provides you with all relevant AI news — without the hype.

Google adds automatic memory and temporary chat controls to Gemini

Google tests Magic View in NotebookLM as potential new data visualization feature

Google releases Gemma 3 270M, an ultra-efficient open-source AI model for smartphones

Claude adds memory search for past conversations

Anthropic expands Claude Sonnet 4 to 1 M token context

Google makes Jules, its AI coding agent, generally available

OpenAI Publishes GPT-5 Prompting Guide and Releases Prompt Optimizer Tool

OpenAI launches GPT-5 with unified fast and reasoning modes for API and ChatGPT

GPT‑OSS: OpenAI Publishes 20B and 120B Open‑Weight Models for Local Deployment

Claude Opus 4.1 Raises Coding and Reasoning Performance

DeepMind Debuts Genie 3 for Real-Time Text-to-3D World Generation

A new generation of world models is here. DeepMind’s Genie 3 turns text prompts into playable 3D environments at 24 fps with persistent memory and interactive elements. While still in research preview, it represents a major step toward AI agents that can learn and act in open-ended virtual worlds.

Alibaba Shrinks Its Coding AI to Run Locally

Qwen3-Coder-Flash is a compact, 30B MoE model capable of local inference on modern MacBooks. It joins Alibaba’s broader Qwen3 ecosystem as a nimble counterpart to the heavyweight 480B hosted version, giving developers a pragmatic hybrid setup for coding workflows.

NVIDIA Releases Llama Nemotron Super v1.5 to Push Open-Source Agent Reasoning

The new 49B model tops open benchmarks with a 128K context window, tool-use capabilities, and single‑GPU efficiency. It's a signal that NVIDIA aims to lead in agent‑focused LLMs that actually run in production.

Krea Releases FLUX.1 Krea Model with Open Weights

The team behind the FLUX image ecosystem just open-sourced FLUX.1 Krea, a distilled model fine-tuned for photorealistic and aesthetic image generation. Developed with Krea, the model is now freely available under a non-commercial license and slots directly into the broader FLUX.1-dev workflow.

Deep Cogito Releases Cogito v2 Models, Challenging Frontier AI on a Budget

Four open-source language models with advanced reasoning and sub-$3.5M training cost aim to rival top-tier AI. Deep Cogito’s approach blends inference-time search with distilled intuition, offering a compelling alternative to closed frontier models.

StepFun open-sources Step3, a 321B parameter VLM optimized for Chinese AI chips

StepFun has released Step3, a massive open-source visual language model with 321 billion parameters and leading benchmark scores. The model debuts with novel attention architectures that reduce inference costs and is optimized to run efficiently on domestic Chinese AI hardware.

Claude Mobile Adds Email and Calendar Actions

n8n Launches No-Setup RAG Template with Vector Store and OpenAI

n8n has released a plug-and-play RAG starter template that lets users upload documents and chat with them instantly. The workflow requires no custom setup and includes preconfigured ingestion and query pipelines using OpenAI.

Ollama Adds Desktop App and Performance Upgrades in v0.10.0

The latest Ollama update brings a native desktop app for macOS and Windows, along with CLI and multi-GPU performance enhancements. It's a move aimed at making local LLM deployment more accessible and efficient across devices.

IBM and Google Launch Rival Toolkits for Document Intelligence

Two enterprise AI contenders just released open-source toolkits aimed at extracting structured insights from unstructured documents. IBM’s Docling and Google’s LangExtract target different parts of the same problem—streamlining how organizations turn documents into usable data.

We love

and you too

Signal vs. Noise

The Pure Neo AI Timeline provides you with all relevant AI news — without the hype.

Google adds automatic memory and temporary chat controls to Gemini

Google tests Magic View in NotebookLM as potential new data visualization feature

Google releases Gemma 3 270M, an ultra-efficient open-source AI model for smartphones

Claude adds memory search for past conversations

Anthropic expands Claude Sonnet 4 to 1 M token context

Google makes Jules, its AI coding agent, generally available

OpenAI Publishes GPT-5 Prompting Guide and Releases Prompt Optimizer Tool

OpenAI launches GPT-5 with unified fast and reasoning modes for API and ChatGPT

GPT‑OSS: OpenAI Publishes 20B and 120B Open‑Weight Models for Local Deployment

Claude Opus 4.1 Raises Coding and Reasoning Performance

DeepMind Debuts Genie 3 for Real-Time Text-to-3D World Generation

Alibaba Shrinks Its Coding AI to Run Locally

Qwen3-Coder-Flash is a compact, 30B MoE model capable of local inference on modern MacBooks. It joins Alibaba’s broader Qwen3 ecosystem as a nimble counterpart to the heavyweight 480B hosted version, giving developers a pragmatic hybrid setup for coding workflows.

NVIDIA Releases Llama Nemotron Super v1.5 to Push Open-Source Agent Reasoning

The new 49B model tops open benchmarks with a 128K context window, tool-use capabilities, and single‑GPU efficiency. It's a signal that NVIDIA aims to lead in agent‑focused LLMs that actually run in production.

Krea Releases FLUX.1 Krea Model with Open Weights

The team behind the FLUX image ecosystem just open-sourced FLUX.1 Krea, a distilled model fine-tuned for photorealistic and aesthetic image generation. Developed with Krea, the model is now freely available under a non-commercial license and slots directly into the broader FLUX.1-dev workflow.

Deep Cogito Releases Cogito v2 Models, Challenging Frontier AI on a Budget

Four open-source language models with advanced reasoning and sub-$3.5M training cost aim to rival top-tier AI. Deep Cogito’s approach blends inference-time search with distilled intuition, offering a compelling alternative to closed frontier models.

StepFun open-sources Step3, a 321B parameter VLM optimized for Chinese AI chips

StepFun has released Step3, a massive open-source visual language model with 321 billion parameters and leading benchmark scores. The model debuts with novel attention architectures that reduce inference costs and is optimized to run efficiently on domestic Chinese AI hardware.

Claude Mobile Adds Email and Calendar Actions

n8n Launches No-Setup RAG Template with Vector Store and OpenAI

n8n has released a plug-and-play RAG starter template that lets users upload documents and chat with them instantly. The workflow requires no custom setup and includes preconfigured ingestion and query pipelines using OpenAI.

Ollama Adds Desktop App and Performance Upgrades in v0.10.0

The latest Ollama update brings a native desktop app for macOS and Windows, along with CLI and multi-GPU performance enhancements. It's a move aimed at making local LLM deployment more accessible and efficient across devices.

IBM and Google Launch Rival Toolkits for Document Intelligence

Two enterprise AI contenders just released open-source toolkits aimed at extracting structured insights from unstructured documents. IBM’s Docling and Google’s LangExtract target different parts of the same problem—streamlining how organizations turn documents into usable data.

We love

and you too

Never miss an update!