Skip to main content
Back

OpenAI ChatGPT Images 2.0: readable text, thinking mode, 2K res

OpenAI just dropped ChatGPT Images 2.0, and it finally fixes the stuff that made AI image tools frustrating to actually use.

April 22, 2026 6 min read

Hey,

AI is quietly crossing a line: models don't just generate anymore, they plan, reason, then execute. OpenAI's Images 2.0 thinks before it draws. Kimi K2.6 codes autonomously for 13 hours straight. A 15M parameter world model plans 48x faster than giants.

The pattern is clear. Raw capability is table stakes now. The real race is autonomous execution over time, not single-shot outputs.

Open-source is keeping pace, which nobody predicted this fast.
Together with · partner_image
Author photo
Today's Author
Lior Alexander. Founder of AlphaSignal and former ML Engineer. Previously built ML systems at Iguazio, Guesty, Enphase, Mila.
Summary

Read time: 6 min 34 sec

Top News

▸ OpenAI ships ChatGPT Images 2.0 with thinking, 2K resolution, and web search

Bright Data

▸ Most agent frameworks break on the real web

Top News

▸ Kimi drops open-source coding model running 300 parallel agents for 12+ hours

Fuel iX by TELUS Digital

▸ Can your LLM withstand a prompt injection attack?

Top Repo

▸ Developer reverse-engineers Claude's design system and open-sources it free

Signals

1. New 15M parameter world model trains on one GPU and plans 48x faster than foundation models
2. Nous Research ships a self-improving open-source agent that learns from every conversation
3. New fine-tuning method teaches LLMs to reason through problems they can't solve directly
4. New 18B frankenmerge model runs on 12–16 GB GPUs by stacking two 9B models
5. New paper warns LLM users mistake AI output for their own real skill
Top News
OpenAI ships ChatGPT Images 2.0 with thinking, 2K resolution, and web search
26,708 Likes
OpenAI ships ChatGPT Images 2.0 with thinking, 2K

OpenAI just dropped ChatGPT Images 2.0, and it finally fixes the stuff that made AI image tools frustrating to actually use.

The old problem? AI images looked cool but broke on real tasks: warped text, ignored instructions, wrong layouts. Images 2.0 targets exactly those weak spots.

Here's what's new and why it matters:

  • Text that actually renders -- readable typography inside posters, infographics, and UI mockups, up to 2K resolution
  • Thinking mode -- the model reasons through your prompt before drawing, searches the web for real-time info, self-checks outputs, and generates up to 8 consistent images from one prompt
  • Aspect ratios from 3:1 to 1:3 -- banners, slides, posters, mobile graphics, all ready to use
  • Multilingual text that flows naturally inside images, not just translated words dropped in

For API access, call gpt-image-2 through the Images API or Responses API. Thinking mode requires Plus, Pro, or Business. Base model is free for all ChatGPT and Codex users.

Read more
forward →
Presented by Bright Data
Most agent frameworks break on the real web

Most agent frameworks assume clean input. The real web doesn't care.

Bright Data Web Scraper API handles JavaScript-rendered pages, bot detection, CAPTCHAs, and rate limits - so you don't have to.

  • One API call → clean JSON output
  • 600+ pre-built scrapers (Google, Amazon, LinkedIn & more)
  • Agent-ready: plug structured data straight into your LLM pipelines

Built for AI products that depend on reliable web intelligence.

Start scraping smarter - 25% off!

Scrape smarter
partner with us →
Top News
Kimi drops open-source coding model running 300 parallel agents for 12+ hours
17,821 Likes
Kimi drops open-source coding model running 300 pa

Moonshot AI just dropped Kimi K2.6, a fully open-source coding model that can work autonomously for over 12 hours straight without asking you a single question.

Think of it like hiring a senior engineer who never sleeps. You give it one prompt, it spins up 300 parallel sub-agents (mini AI workers), and they all chip away at your project simultaneously. That's triple the agents K2.5 could handle.

Here's what it actually pulled off in testing:

  • Rewrote a financial matching engine over 13 hours, made 1,000+ tool calls, and boosted throughput by 185%
  • Built a local AI runtime in Zig (a rare systems language) from scratch, hitting 193 tokens/sec
  • Generates full frontends with animations, auth, and databases from one prompt

The API is OpenAI-compatible at $0.60 per million input tokens, roughly a quarter of Claude Opus pricing. Weights are on Hugging Face, and you can run it locally via Ollama.

Read more
forward →
Presented by Fuel iX by TELUS Digital
Can your LLM withstand a prompt injection attack?

We tested 24 leading GenAI models for vulnerabilities. All 24 failed, with attack success rates up to 64%.

Every model demonstrated exploitable flaws that turn AI into an attack vector.

Download our report for security profiles on all 24 models and strategies to secure your AI stack.

Access the report
partner with us →
Top Repo
Developer reverse-engineers Claude's design system and open-sources it free
3,774 Likes
Developer reverse-engineers Claude's design system

Someone reverse-engineered Claude's design system and turned it into a free, open-source skill called Huashu Design. Think of a "skill" as a plugin you drop into your AI coding agent, and it instantly gains new abilities.

One command gets you started:

npx skills add alchaincyf/huashu-design

Works with Claude Code, Cursor, Codex, and more. You just describe what you want in plain text, and the agent builds it. Here's what you can ship:

  • Clickable iOS/web app prototypes in a real iPhone frame, auto-tested with Playwright
  • Presentation slides that export as editable PPTX (real text boxes, not images)
  • Animations exported as MP4 or GIF, with background music included
  • A 5-dimension design review with a radar chart and a fix list

The coolest part: it pulls real brand colors from official sources instead of guessing. No Figma, no GUI, no buttons. Just type, hit enter, get a finished design.

Read more
forward →
Signals
1
New 15M parameter world model trains on one GPU and plans 48x faster than foundation models
3,248 Likes
2
Add live speech translation to your app in minutes. Streaming API, <1s latency, 60+ languages, Python/JS/Java SDKs.
Presented by Palabra
3
Nous Research ships a self-improving open-source agent that learns from every conversation
109,020 Stars
4
New fine-tuning method teaches LLMs to reason through problems they can't solve directly
Must Read
5
New 18B frankenmerge model runs on 12–16 GB GPUs by stacking two 9B models
28,638 Downloads
6
New paper warns LLM users mistake AI output for their own real skill
1,044 Likes
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We're passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.

Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We're based in the United States. By using our site, you agree to be governed by U.S. laws.
Looking to promote your company, product, service, or event to 250,000+ AI developers?
Work with us