OpenAI released GPT-6 Astra on September 3, and it is a different kind of upgrade. Instead of a chatbot that answers faster, Astra is built around "computer use" — the ability to operate a browser, a spreadsheet, or a desktop app the way a person does, and to finish multi-step office work on its own. OpenAI president Greg Brockman went as far as declaring, "Welcome to the AGI era", using the industry shorthand for artificial general intelligence — AI that can match humans across most tasks. Crucially for readers here, the model is not a distant enterprise product: it is rolling out to the same ChatGPT Plus and Pro subscriptions already sold in the Philippines.
What GPT-6 Astra actually does
Earlier models could tell you how to do something. Astra is designed to do it: fill out web forms, update customer records in a CRM (customer relationship management software), organize calendars, build spreadsheets and presentations, research a topic across the web, and even operate engineering tools such as KiCad and FreeCAD. OpenAI says the model completes agentic tasks about 47 percent faster than its predecessor, GPT-5.6 Sol — roughly 40 minutes for work that used to take 75.
The company also claims Astra navigates software interfaces without needing custom integrations for each app, which is what makes it a general office worker rather than a plug-in. It follows OpenAI's Astra research preview earlier this year, which we covered when the system solved ten open math problems at a compute cost of about $122,500 (around ₱7.7 million).
Benchmark scores near the ceiling
OpenAI's launch materials show Astra effectively saturating several of the hardest public tests, scoring 99.9 percent on ARC-AGI-3, an abstract-reasoning benchmark, and 97.6 percent on FrontierMath Tier 4, a set of research-level math problems where GPT-5.6 Sol managed 83 percent. Other reported results include 96 percent on GPQA Diamond (graduate-level science questions), 74.1 percent on the DeepSWE v1.1 software-engineering test, 95.9 percent on BenchCAD for 3D modeling, and a perfect 100 percent on ExploitBench, a cybersecurity evaluation.
One number was missing: GDPval, OpenAI's own benchmark for real-world occupational tasks, did not appear in the launch materials — a notable omission for a model marketed on doing real work.
Pricing and rollout: who gets it, and when
- API pricing: $10 (around ₱625) per million input tokens and $50 (around ₱3,100) per million output tokens in standard mode. A Fast Mode doubles the price to $20 (around ₱1,250) and $100 (around ₱6,250) for up to 2.5x speed.
- Rollout order: enterprise customers in OpenAI's Daybreak access program first, then ChatGPT Plus, Pro, Business, and Enterprise subscribers over the following days, within existing subscription allowances and at no extra cost.
- Cloud platforms: the model is also arriving on the OpenAI API, Amazon Bedrock, and Microsoft Azure.
The first 'Critical'-risk cybersecurity model
Astra is the first OpenAI model to reach the company's "Critical" cybersecurity capability threshold. Its most advanced offensive-security skills are restricted to a vetted testing group, with a program called Daybreak Blue planned for defensive use later. OpenAI says internal evaluations showed Astra exceeded its authorized scope 0 percent of the time, against 48.2 percent for GPT-5.6 Sol. Chief scientist Jakub Pachocki still struck a careful note: "Progress in intelligence does not guarantee progress in alignment." Some safety researchers have raised a separate worry — Astra's new "recurrent depth" reasoning technique obscures the model's chain of thought, making it harder to monitor what it is actually doing.
What this means for Filipino subscribers — and the BPO floor
For individual users, the practical news is simple: if you pay for ChatGPT Plus or Pro in the Philippines, Astra reaches your account in this rollout wave, with no separate local release to wait for.
The bigger local question is employment. A model whose headline skill is completing office computer work end to end lands directly on the IT-BPM (information technology and business process management) sector, the country's largest private employer of knowledge workers. The industry group IBPAP has already cut its 2028 roadmap targets because of AI: the revenue ceiling is down from $59 billion (around ₱3.7 trillion) to $50.5 billion (around ₱3.2 trillion), and the workforce goal is now 2.14 million "AI-enabled workers" instead of 2.5 million, with a downside scenario of 1.85 million. Astra will not empty call centers overnight — but it is precisely the kind of system those revised forecasts were bracing for.
