Comply with ZDNET: Add us as a most popular supply on Google.
ZDNET’s key takeaways
- OpenAI’s GPT-5.6 launch appears aimed instantly at Anthropic.
- Sol, Terra, and Luna map intently to GPT-5.5’s tiers.
- ChatGPT Work brings OpenAI’s agent instruments past coding.
OpenAI cranked out a sequence of huge bulletins at present. They’re news-making in their very own proper, however I’ve to be sincere with you. These bulletins additionally really feel like they’re OpenAI’s means of enjoying catch-up with all the excitement Anthropic has been producing.
Key among the many bulletins are three new fashions: GPT-5.6 Sol, Terra, and Luna. We knew that 5.6 was on the way in which as a result of the federal government stepped in final month to delay its launch. This, in fact, follows scorching on the heels of the massive news-dominating Anthropic/White Home kerfuffle about Fable 5 and Mythos 5.
Also introduced was ChatGPT Work, which is clearly OpenAI’s reply to Claude Cowork. ChatGPT Work permits the AI to entry your desktop and browser and carry out duties for you.
This isn’t a brand new OpenAI function. Beforehand, it was embedded within the Codex app, OpenAI’s agentic programming software. Clearly, OpenAI took a branding web page from Anthropic and separated the agentic work assistant from its agentic coding software, which ought to make it extra interesting to non-coders.
I’ve typically criticized OpenAI for its startling absence of an excellent branding technique, so that is really a constructive signal. Maybe its programmers are now not driving their product advertising technique. Then once more, maybe ChatGPT 5.6 Sol is.
Benchmarks put Claude comparisons within the announcement
There is no doubt OpenAI has Anthropic in its sights. The corporate’s announcement of the brand new fashions instantly targets testing outcomes and pricing ranges of Anthropic’s choices. However first, let’s do a fast rundown of these celestial mannequin names.
- Sol: That is the flagship mannequin, aimed straight at Fable 5. It’s roughly equal to GPT-5.5’s pondering mode.
- Terra: Because the Earth is smaller than the solar, Terra is much less succesful than Sol. Terra is the default, mainstream GPT-5.6 mannequin, equal to GPT-5.5.
- Luna: Because the moon is smaller than the Earth, Luna is much less succesful than Terra. Consider it because the equal of GPT-5.5’s immediate mannequin.
Brokers’ Final Examination is a benchmark for testing how nicely AI fashions deal with long-running, multi-step skilled workflows throughout 55 fields. OpenAI says GPT-5.6 Sol scored 53.6 on Brokers’ Final Examination, beating Claude Fable 5 by 13.1 factors.
Synthetic Evaluation Intelligence Index is one other benchmark. It makes an attempt to quantify broad mannequin intelligence throughout a number of classes. OpenAI says Sol got here inside one level of Fable 5 whereas finishing duties in 61% much less time.
In one more benchmark, OpenAI says GPT-5.6 Sol reached 80 on the Synthetic Evaluation Coding Agent Index, 2.8 factors above Fable 5. Are you seeing a theme right here?
OpenAI additionally says that GPT-5.6, and particularly Sol, is rather more environment friendly and due to this fact more cost effective than Fable. At medium reasoning, OpenAI says Sol beat Fable 5 by 11.4 factors at roughly one-quarter the estimated price.
GPT-5.6 can be sooner and extra environment friendly than GPT-5.5. Qodo is an AI code evaluation software that offers groups deep codebase context to create sooner and enhance code high quality.
Itamar Friedman, Qodo’s co-founder and CEO says, “GPT‑5.6 was the strongest mannequin we evaluated on our agentic code-review exams. On our apples-to-apples inside and exterior PR benchmarks, it beat GPT‑5.5 on F1 whereas utilizing roughly 3x fewer tokens per PR and delivering about 2x decrease median latency.”
In spite of everything the federal government fuss about mannequin security and misuse, OpenAI says GPT-5.6 consists of safeguards that mix mannequin protections, real-time checks, monitoring, and entry calibrated to belief and threat. Based on the announcement, “GPT-5.6 launches with our most sturdy safeguards thus far.”
ChatGPT Work
After which there’s ChatGPT Work, OpenAI’s agentic desktop productiveness software. Suppose Claude Cowork, however with OpenAI’s fashions. ChatGPT Work can create spreadsheets, PowerPoints, and full web sites. It will possibly work on complicated tasks for hours by breaking them into smaller duties.
In one of many scariest demos I’ve seen in a very long time, an OpenAIer let ChatGPT Work free in his Apple Notes, permitting the software to utterly reorganize all his notes. Now, I gotta let you know. I closely depend on the notes in my Apple Notes, and the concept the AI would go in and transfer all that stuff round freaks me out.
It is a undoubtedly highly effective instance of what ChatGPT Work can do, however I’ve belief points at this degree of delegation.
Based on OpenAI, “One of the best ways to learn to use ChatGPT Work is to offer it a process you already know nicely: analyze a month-end price range variance, flip supply supplies right into a advertising marketing campaign transient, or put together for a gross sales assembly. You possibly can observe its progress, reply questions, change path, and approve vital actions.”
I have been utilizing Claude Cowork to do a few of these bigger, extra complicated agentic workflows, and it is saved me loads of time. Whereas I have never but examined ChatGPT Work, early customers report comparable advantages.
Angela Ferrante, head of enterprise advertising at Zapier, is quoted within the announcement as saying, “[We] used ChatGPT Work to construct a repeatable system for reviewing 1000’s of leads every month. It traced buyer touchpoints throughout Zapier’s CRM, electronic mail, and different instruments, discovered the place follow-ups broke down, and generated a weekly govt dashboard that highlighted missed pipeline and revealed seven figures in potential gross sales.”
In the identical means that Claude Cowork connects to exterior purposes, ChatGPT Work has plugins and connectors that work with Slack, Microsoft Groups, Google Drive, SharePoint, electronic mail, calendars, CRMs, and undertaking trackers, amongst different instruments.
Availability
ChatGPT Work is rolling out on net and cell for Professional, Enterprise, and Edu plans, with Plus and Enterprise following over the following few days. I am inquisitive about how a lot you’ll get performed with Work on the $20/mo Plus plan. Since that is the tier I am spending on now, I will be testing it out, and I am going to report again to you.
Up till now, OpenAI had separate ChatGPT and Codex desktop apps. They’re combining the 2 into one app. Anthropic combines Claude chat, Claude Code, and Claude Cowork into one desktop app. From the announcement, it wasn’t clear whether or not ChatGPT Work will probably be folded into the newly mixed ChatGPT/Codex app. This, too, I am going to let you know extra about as soon as I am going hands-on.
Would GPT-5.6’s sooner efficiency and decrease price make you extra possible to decide on OpenAI over Anthropic? Tell us within the feedback beneath.
You possibly can observe my day-to-day undertaking updates on social media. Make sure you subscribe to my weekly replace e-newsletter, and observe me on Twitter/X at @DavidGewirtz, on Fb at Fb.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.
