OpenAI Releases ChatGPT Images 2.0, Claiming Top Spot
OpenAI is back at the forefront of AI image generation with its latest release, ChatGPT Images 2.0. The company calls it the “smartest image generation model ever built,” and the results support the claim. This is not just an incremental update; it is a fundamental shift in how AI creates visuals.
The model's standout feature is its ability to 'think' before generating. It can plan its output, search the web for real-time context and references, and even check its work for errors before delivering the final images. This leads to visuals with higher fidelity and better instruction following.
Key upgrades include improved text rendering, multi-image reasoning, and support for high-resolution 2K outputs. It can generate up to eight images at once across various aspect ratios, from ultrawide 3:1 to tall 1:3. According to the popular Arena AI leaderboard, ChatGPT Images 2.0 has decisively swept every category, dethroning the long-standing leader, Nano Banana 2. OpenAI CEO Sam Altman described the leap as being “like going from GPT-3 to GPT-5 all at once.” The model is now live in ChatGPT, Codex, and via the API.
Google and Anthropic Push Agent Technology Forward
While OpenAI dominates the image space, competitors are making significant strides in agentic AI. These are systems designed to perform tasks autonomously, and recent developments show they are becoming increasingly sophisticated.
Google's Deep Research Max
Google has introduced Deep Research Max, a state-of-the-art agent powered by the Gemini 3.1 Pro model. It is designed to generate comprehensive research reports by accessing the web, uploaded files, or private data through a Model Context Protocol (MCP) server. Google's internal benchmarks show significant performance jumps over models like Opus 4.6 and GPT 5.4 in retrieval and reasoning tasks.
The Rise of Always-On Agents
The concept of persistent AI assistants is gaining traction. OpenAI is reportedly developing an always-on agent platform codenamed Hermes within ChatGPT, allowing users to create and run custom agents that perform scheduled tasks without constant prompting. Similarly, Anthropic is working on its own agent, codenamed Conway, which features UI extensions for managing tasks and integrations across web and mobile.
Specialized Tools for Developers and Security
The AI tooling landscape is also diversifying with highly specialized models and platforms for technical users. These tools are moving beyond general-purpose chat to solve specific, complex problems in cybersecurity and software development.
Anthropic's Claude Mythos Finds Decades-Old Bugs
Anthropic's Claude Mythos is a powerful cybersecurity model that is not available to the public due to its potential for misuse. As part of a private initiative called Project Glasswing, it was used to help Firefox patch an incredible 271 security vulnerabilities in a single release. The model demonstrated its power by identifying a 27-year-old bug in OpenBSD and a 16-year-old flaw in FFmpeg that had evaded scanners for years.
New Open-Source and Agentic Tools
Several other notable tools have been released for developers. Moonshot AI launched Kimi K2.6, a new open-source coding model for complex engineering tasks. Exa released Deep Max, a new agentic search tool that it claims is 20x faster than rivals. For security, CrabTrap is a new open-source HTTP proxy that uses an LLM-as-a-judge to secure AI agent requests. Meanwhile, Google open-sourced its DESIGN.md format from its Stitch tool, allowing agents to understand a project's design system.
A Roundup of New and Niche AI Applications
The industry continues to see a rapid launch of new tools for various use cases. Here are some of the latest we have been tracking:
- Qwen3.5-Omni: A large-scale multimodal model that natively processes text, audio, images, and video in a single architecture.
- Genspark Build: An agentic tool powered by Claude Opus 4.7 that generates entire applications and websites from text prompts.
- Business & Productivity: New tools include WorkOS for auth integration, Dageno for tracking brand mentions in AI, Devaito for website and automation building, Viktor as a Slack-based AI assistant, and PopTask for task management.
- Content & Media: We're seeing new tools like Scrunch for AI site audits, Relumi for photo restoration, Inkfluence for creating ebooks, and EasyScribe for transcription and subtitling.
- Skills Assessment: AISA offers a unique way to assess your AI skills through a live conversation with two different AIs.
