GPT-6 Astra Put to the Test on a 90-Day Business Plan

OpenAI claims Astra is the strongest model so far for complex work. Several model updates are worth knowing about:

• It handles unclear instructions better. Astra is better at asking questions when missing information could affect the result.
• It keeps a better track of the original task. When you add a requirement or ask a side question, it responds better without forgetting the goal or earlier instructions.
• It follows your template and style more closely. It finds relevant information in existing documents, slides, and spreadsheets. This rich context allows Astra to respond better to your existing format.
• It uses computer tools more effectively. With access, Astra can interpret screens and make more efficient, accurate decisions about where to click or what to type.

Additional platform updates enhance Astra’s performance. This includes:

• More approval checkpoints and confirmation policies in Enterprise workspaces. Admins can require approval before a consequential action goes through. OpenAI also automatically reviews tool calls that look unsafe or unauthorized.
• OpenAI models can work inside apps without an API. No need to wait for an official plugin, as Astra opens the app itself, reads what's on the screen, and clicks and types through it the way a person would. On the Enterprise plan, admins can limit that to an approved list of websites and apps.

Who has this today? It's live now inside ChatGPT on Plus, Pro, Business, and Enterprise, plus Codex and the API.

On Enterprise, an admin has to switch it on for the whole workspace and authorize sites and apps Astra can access.

I gave five options in the Model selection toggle (GPT-5.6 Terra Light, GPT-5.6 Sol Light, GPT-5.6 Sol Medium, GPT-6 Astra Light, and GPT-6 Astra Medium) the same prompt on a strategic business issue: build a 90-day plan, and choose the three top moves worth prioritizing.

The task asked each model to weigh competing opportunities and commit to a business direction.

I temporarily turned off conversation history and deleted each test conversation before starting the next to avoid carryover between tests.

AI responses are non-deterministic, meaning they can vary each time under the same prompt. As such, this assessment is relative and should be paired with long-term user experience for a reliable judgment of which model is worth employing depending on user needs.

The clearest difference was how each model can be deployed for different situations. Terra took a more general approach. Sol gave a more detailed response, while Astra delivered a more compact answer that was still insightful.

Both Sol responses explained the choices step by step. This made the recommendations easier to discuss, question, and further develop.

Sol Light’s diagram added a useful visual overview, while Sol Medium went deeper into details that were easier to miss in the shorter responses.

Astra’s was distinctively compact while still being thorough.

Astra Medium brought the most useful combination: weighted criteria to explain how it ranked the options, enough detail to understand the recommendations, and embedded links to check the supporting material as you read.

Astra’s advantage here was the combination of detail, references, and concision.

GPT-6 Astra, Sol, and Terra can all produce a useful 90-day plan. The difference is the level of detail they use to present the work.

• Use Astra when you need a concise decision brief. Astra Medium gave us the clearest combination of ranking criteria, detailed recommendations, and linked sources. The responses are comprehensive without being too long.
• Use Sol when you need to work through the plan in detail. Its answers explained the logic step by step, surfaced smaller execution considerations, and, in Sol Light’s case, added a helpful visual roadmap.
• Use Terra for a fast brainstorming session. It gave a solid general outline, which can be enough when you need to structure the discussion before developing the plan further.

Depending on the context of your decisions, each model can be useful in different ways.

Better choices come from experimentation to further understand each model’s capabilities.

So try GPT-6 Astra today and see what it has to offer.