AI is getting more useful, and less forgiving

TLDR

- OpenAI says enterprise AI now makes up more than 40% of its revenue, a clear sign that buyers are moving past pilot mode.

- Google and OpenAI both shipped upgrades that make research and image workflows easier to use.

- Anthropic and Meta pushed the conversation forward from two sides, cyber risk and consumer scale.

This month's briefing

1. Enterprise AI is no longer a side budget

OpenAI says enterprise products now represent more than 40% of revenue, up from 25% in late 2025. The same post says Codex reached 3 million weekly active users. That is a strong signal that AI is becoming operating infrastructure, not side tooling.

Source: OpenAI, Next phase of enterprise AI

2. Anthropic put cyber capability at the center of the month

Anthropic's technical write-up on Claude Mythos Preview is one of the clearest signals this year that stronger coding systems are converging with offensive cyber workflows. Even if you never touch security research, it changes how seriously teams need to treat access, logging, and permission boundaries.

Source: Anthropic, Claude Mythos Preview

3. Meta pushed a new large-scale assistant model into the conversation

Meta introduced Muse Spark and says it currently powers the Meta AI app and website, with rollout to WhatsApp, Instagram, Facebook, Messenger, and AI glasses coming in the following weeks. The bigger point is distribution: consumer AI keeps getting built into products people already use all day.

Source: Meta, Introducing Muse Spark

4. Google made deep research more practical

Google's next-generation Gemini Deep Research now includes native spreadsheets, visualizations, a stronger Gemini 3.1 Pro model, and Deep Research Max for higher-end work. Google also says it supports the Model Context Protocol, which matters if your research stack needs cleaner links to external tools and company knowledge.

Source: Google, Next-generation Gemini Deep Research

5. Consumer AI creation got easier this month too

OpenAI's April 21 release notes say ImageGen 2.0 is now available to all ChatGPT plans, while Thinking remains for paid users. That matters because image generation is no longer tucked away in one premium workflow. It is becoming normal product and content work.

Source: OpenAI Help Center, ChatGPT release notes

THE NUMBER

3 million

That is the number of weekly active Codex users OpenAI reported on April 8, 2026. The more interesting signal is what surrounds it: enterprise AI is turning into repeat usage inside teams, not occasional experimentation.

Source: OpenAI, Next phase of enterprise AI

EDITOR'S PICK

Google Deep Research Max

If your work involves market scans, vendor comparisons, or briefing notes, this is one of the most practical launches of the month. The Max tier, stronger Gemini 3.1 Pro model, native spreadsheets, and visualizations make it easier to go from question to structured draft without juggling three tools first.

ONE THING TO TRY

Run one real task with a two-step research workflow this week:

1) Ask a research system to build the first pass, with sources and a table.

2) Ask a second pass, human or model, to check every number, link, and claim against the original sources.

April's launches make step one faster. April's security warnings are a good reminder that step two still matters.

YOU ASKED, WE ANSWERED

"What actually changed in AI image tools this month?"

The short answer is access. On April 21, OpenAI said ImageGen 2.0 became available across all ChatGPT plans. That matters because image generation is moving from premium add-on to normal day-to-day workflow for social posts, product mockups, campaign concepts, and internal drafts.

Builder tip: keep a tiny prompt library for repeatable visual tasks like hero image drafts, ad concepts, or thumbnail variants. Consistency usually matters more than novelty.

Source: OpenAI Help Center, ChatGPT release notes

IF YOU READ ONE THING

Anthropic's Mythos Preview write-up

This is the best single technical read from April if you want to understand why the conversation around capable coding systems is shifting. It is not light reading, but it clearly explains why safety, access control, and auditability now belong in product thinking, not just security planning.

SPONSORED SPOTLIGHT

Oniichan

Oniichan is a text-to-manga and anime creation tool focused on character design, reusable character sheets, and more consistent multi-page storytelling. It also supports speech bubbles and comic-style page workflows, and it is free to start.

Explore Best-AI.org

Domain switch. Nothing to worry about: We now run on best-ai.org. The old address, best-ai-tools.org, still works and all links lead to the same place. We are not collecting any data twice. Your profile, favorite lists, and saved tools are all preserved.

For security, you may have been logged out. Log in again and you should see your lists as usual.

Built an AI tool?

Submit it to the directory and join thousands of creators on best-ai.org.