What happened
Tibor Blaho reports that GPT-6.2 Astra and Sol handled tasks ranging from server management to code review in his day-to-day work. In a test of Astra, he upgraded a multimillion-line legacy application and its dependencies from an outdated programming language version; it was usable locally after about three hours, and all tests passed from a clean checkout after about six hours. He says he plans to keep using Astra and move other tasks to Sol.
Why it matters
Astra upgraded a multimillion-line legacy application to the current language version, with tests passing from a clean checkout after about six hours, showing how far the author's real-world migration task progressed.
I've been testing GPT-6.2 Astra and Sol on my day-to-day work
Most of what I gave them, small or bigger tasks, they got done - finding out why something broke, managing servers across multiple data centers, building new projects, fixing defects, reviewing changes, taking care of backups and monitoring
Also smaller tasks like research, admin and office work
During testing of GPT-6 Astra, it moved a nearly 20-year-old codebase from one version control system to another including the complete history
This time I asked the new Astra to upgrade that same multimillion-line legacy app, with its legacy dependencies, from a very outdated version of its programming language to the current one
After about 3 hours I could already use it locally, and after about 6 hours all tests passed from a clean checkout
There's of course still more work to do, but the first results are very promising
I used them mostly in the Codex app and in T3 Code, which I now prefer for most of my work
T3 Code runs on my Mac, and I can connect to it from my Omarchy machine or from the iOS app over VPN and continue where I left off
I use Codex and Claude Code directly from T3 Code, so I can hand off between them and orchestrate subagents across both, because I use Astra together with Fable & Opus
I let one do the research and give its findings to the other to see what it missed, to brainstorm how to fix something or how to approach a problem
They also review each other's work - Astra still finds defects in Fable's code and Fable still finds defects in Astra's code
So it still makes sense to use both in my opinion, for me Astra and Fable are the best team currently available
Every new model works and communicates a bit differently than the one before, and so far adapting how I use it was enough for me to work well with it
Pro 200 gets Ultrafast for the first time (with Sol), and Pro 500 gets it with both Astra and Sol
Astra is also 20% cheaper in the API now - $8 input and $40 output per 1M tokens
For my work these are good models - I'll keep using Astra and move my other tasks to the new Sol
View the quoted post on XSome of you have been asking very politely. Some of you have just been asking.
GPT-6.2 Astra and Sol are now available.
Super efficient and built to perform where it counts—the real world.
The story so far
OpenAI released the GPT-6.2 Astra and Sol models, which are available now in ChatGPT Work, Codex, and the API. Astra, positioned as the company’s new flagship model, offers reduced pricing with API rates 20% lower than its predecessor at $8 per million input tokens and $40 per million output tokens, while Sol maintains the existing $2 and $10 pricing structure. OpenAI also expanded access to its Ultrafast mode to include Pro 200 and Pro 500 subscription tiers. The company reports that GPT-6.2 Astra sets new highs on benchmarks including Agents' Last Exam, BenchCAD 2.1, and Terminal-Bench Science 0.1 at lower task costs. However, external evaluations show mixed results. Epoch AI reported that GPT-6.2 Astra scored 65% on internal automation quality standards, trailing Fable 5.2. Additionally, independent testing by Simon Willison indicated that the model struggled with animation tasks compared to other frontier models. OpenAI also implemented tighter cybersecurity safeguards for the new models and plans to provide access through its Daybreak platform for approved defenders, though the company noted that these safeguards may increase refusal rates for some security research tasks.
How the story developed · 3
View full event- 11/18 06:15 (UTC+8)
Tibor Blaho Tests GPT-6.2 Astra and Sol on Daily Work This report
- 11/18 06:15 (UTC+8)Perplexity Shares GPT-6.2 Astra and Sol WANDR EvalsAravind Srinivas · X
- 11/18 02:36 (UTC+8)OpenAI Introduces GPT-6.2 Astra and SolOpenAI News
