GPT-6 Astra scores 99.9% on ARC-AGI-3, OpenAI says
OpenAI has released GPT-6 Astra, a model that follows GPT-5.6 Sol. According to OpenAI's official announcement, the new model achieved a 99.9% score on the ARC-AGI-3 benchmark, a significant jump from the 7.8% recorded by its predecessor. This performance has led some observers to speculate about the arrival of AGI, though such claims remain unverified.
What you can do with it
AI at Work →Check subtitles and captions for errors in videos before client review
What you get You can review short-form videos and spot errors in long-form edits.
Analyzes video text against audio and visuals.
- Who for
- marketers
- Cost
- Price not stated
Use it for
- Review short-form videos
- Spot errors in long-form edits
- Draft project proposals faster
How to set it up · From the article
- Upload or paste video transcript and subtitles into Codex’s GPT-6 Astra
- Ask for timestamped discrepancies, typos, and stylistic issues
- Manually verify AI flags against the original video
- Apply corrections in your editing software before client review
Watch out Results depend on input quality
Key points from the news
- OpenAI reports GPT-6 Astra scored 99.9% on ARC-AGI-3, up from 7.8% for GPT-5.6 Sol.
- User created a 3D model, game, and website with single prompts using Max inference settings.
- Reset tickets allow users to reset weekly usage limits, reducing need for multiple accounts.
Mr. Nagashima, a Japanese AI educator and operator of the Web Shoku TV channel, demonstrated seven practical use cases for the model in a recent video. He created a 3D model in Blender, Canva slides, a Mario-style game, and a business management system using single-word or minimal prompts. Notably, the model used computer-use capabilities to operate Canva and called GPT Image 2 directly to generate game graphics without external API keys.
Nagashima reported that setting the inference volume to 'Max' significantly improved visual quality. He also highlighted the utility of 'reset tickets,' which allow users to reset weekly usage limits at their discretion, reducing the need for multiple accounts. He noted that the model's usage volume is several times higher than that of Anthropic's Fable 5.1. The video also mentioned the development environment Orca, used to manage parallel coding sessions with Codex and Claude Code.
Model pages: GPT-6 Astra → · GPT-5.6 Sol →
The story so far
2 episodes →- GPT-6 Astra scores 99.9% on ARC-AGI-3, OpenAI saysthis story
7 Deliverables Created by Asking GPT-6 Astra for a Single Word, and Insights on Settings and Limitations After Using It for a Few Days
note.com · 25 September 2026
Loading the full article…
This text was published by note.com and written by TAKERO. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
6sources- Reddit discussionreddit.com
- OpenAI's GPT-6 Astra can now tell you exactly where you screwed up your IKEA shelfPress · The Decoder ·
- The Buzz Around 'GPT-6 Astra': How to Use It for Side Hustles? From 'AI That Answers' to 'AI That Works With You'Press · note.com ·
- Will AI make video editors obsolete? The most surprising thing I discovered using Codex's GPT-6 Astra for work.Press · note.com ·
- How invideo improves color grading 3x with GPT‐6 AstraPrimary source · OpenAI ·
- Astra leads in IKEA furniture assemblyPress · epoch.ai ·
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Apiiro finds 1,122 public GitHub files with secrets saved by Claude Code · 1 src
- Thinking Machines releases Inkling 2 Small, a 594B MoE model with 17B active parameters · 8 src
- OpenAI releases GPT-6 Cyber, a Critical-rated model that outperforms predecessors in vulnerability detection and alignment · 25 src
- Google releases Gemini Omni 1.2 Flash with 15-second video generation and 30% price cut to $0.07 per second · 10 src
- US accuses Chinese executive of allegedly sharing Nvidia chip info with Beijing · 11 src
Comments
via GitHub Discussions