Mistral releases Mistral Small 4 model under Apache 2.0 licence

Mistral has released Mistral Small 4 under Apache 2.0, a unified open-source model combining instruction following, deep reasoning, and multimodal capabilities. With 119B parameters, 256k context, lower latency, and higher throughput, it targets scalable enterprise and research deployment.

🗞 #mistral

Mistral and Nvidia join forces to co-develop open-source models

Mistral AI joins Nvidia’s Nemotron Coalition to co-develop open frontier models, launching Mistral Small 4 with a 128-expert MoE architecture and 256k context window. The partnership pairs open models with Nvidia’s cloud, chips, and deployment stack for production use.

🗞 #mistral

Google Cloud previews G4 VMs using NVIDIA Blackwell GPUs

Google Cloud introduced fractional G4 VMs powered by NVIDIA RTX PRO 6000 Blackwell GPUs, offering 1/2, 1/4, and 1/8 slices for AI and graphics workloads. The move targets agentic AI teams and precedes the planned rollout of Vera Rubin NVL72 systems in late 2026.

🗞 #ai

ICYMI: Perplexity Computer is now available on Android too!

Besides that, Perplexity Computer on Comet browser can operate open tabs and seamlessly access necessary context.

Computer everywhere 👀

Google is rolling out Personal Intelligence in AI Mode across US users.

The feature requires an explicit OPT-IN ✅

Gemini in Chrome is rolling out more broadly as well!

Perplexity announced Comet Enterprise, a browser with a security layer powered by CrowdStrike, enterprise-grade access controls, integrations, and automation features.

I need it on my 9to5 👀

BREAKING 🚨: OpenAI released GPT-5.4 Mini and GPT-5.4 Nano on the APIs. A mini version is also available on ChatGPT and Codex apps.

Check intelligence 👀

Midjourney V8 is rolling out on Midjourney Alpha!

- V8 is much better at taking detailed directions.
- V8 renders up to 5x faster than V7
- New 2K Mode

Anthropic is rolling out a new Dispatch feature for Claude Cowork that lets users communicate with the Claude Desktop app from their phones!

Currently available to Max subscribers in research preview, with a later expansion to Pro plans.

Mobile Cowork! 👀

Google plans to roll out a new "Vibe Design" experience on Stitch this Wednesday.

Voice mode with selectable voices, a new canvas layout, a new "Hatter" agent, and a lot more are coming.

Stitching... 👀

Mistral AI announced Forge, a new system that enables enterprises to build frontier AI models grounded in their data.

Mistral is so good at naming 👌

OpenAI updated ChatGPT model selector on both web and mobile, allowing users to seamlessly switch between Instant and Thinking modes.

A separate "configure" menu is now available with an extended model selector.

Perplexity launches Comet Enterprise with advanced security

Perplexity has launched Comet Enterprise, a managed AI-native browser for companies that embeds research, task automation, and cross-site workflows directly into tabs. It adds centralized admin controls, policy enforcement, and compliance features for secure deployment.

🗞 #perplexity

OpenAI launches GPT-5.4 mini and GPT-5.4 nano on APIs

OpenAI launched GPT-5.4 mini and nano to deliver faster, lower-cost performance for coding copilots, subagents, and high-volume systems. Mini doubles GPT-5 mini’s speed, while nano targets lightweight tasks, extending GPT-5.4 across API, Codex, ChatGPT, and partner clouds.

🗞 #chatgpt

Mistral AI launches Forge for building AI enterprise models

Mistral AI has launched Forge, a system that enables enterprises to build AI models trained on proprietary data. Supporting multimodal inputs and advanced architectures, it allows secure customization, compliance alignment, and control for regulated and knowledge-intensive sectors.

🗞 #mistral

Cursor trained Composer to self-summarize through RL

Cursor’s Composer is an AI model for agentic coding, trained with reinforcement learning to self-summarize and retain context across long tasks. It cuts compaction errors by 50%, uses ~1,000-token summaries, and outperforms prompt-based methods on complex, multi-step coding benchmarks.

🗞 #cursor

ICYMI 👀: Manus AI now supports Google Workspace CLI and can perform new actions in Docs, Slides and Sheets.

MiniMax launches M2.7 model on MiniMax Agent and APIs

MiniMax has launched M2.7, advancing its M2-series with agent-driven self-improvement, multi-agent collaboration, and 97% skill adherence across 40+ tasks. It delivers strong benchmark results, autonomous debugging, and robust support for software, research, and office workflows.

🗞 #minimax

BREAKING 🚨: MiniMax released MiniMax M2.7, a new self-evolving model, achieving a score of 56.22% on SWE-Bench Pro.

M2.7 was used for building complex agent harnesses during its own development. Users can now access MiniMax M2.7 via APIs and MiniMax Agent.