OpenAI (GPT-6 Astra / GPT-5.6 family): the default school-platform provider, and why three model types matter

Most Hong Kong teachers meet OpenAI through ChatGPT first, yet the common procurement mistake is assuming "good enough for chat" means "good enough for a school platform" — OpenAI's value to a school is that one vendor supplies chat, embedding and speech models together.

OpenAI is Edor.ai's default native provider and the AI brand teachers encounter first. For school procurement its significance is not "newest and strongest" but that one API key can support chat, knowledge-base retrieval and oral practice at once — something vendors that make language models only cannot replicate.

From consumer product to school infrastructure

OpenAI reaches the outside world through three product lines, and procurement needs to keep them separate:

• ChatGPT (consumer): fine for an individual teacher planning a lesson, but no student safety gating, no teacher monitoring, and no school ownership of the data.
• OpenAI API (developer): the route Edor.ai uses, billed by token, with no-training terms available.
• Azure OpenAI: the same models, managed inside a school or sponsoring body's Azure subscription, suited to schools already on Microsoft contracts.

Most people know only the first. What a school platform actually depends on are three model types behind the API: language models (the GPT family) for planning, question-setting, marking and student conversation; embedding models that turn school policies and teaching materials into vectors — the foundation of a knowledge base; and Whisper plus TTS for Cantonese, Putonghua and English oral practice. Anthropic and Poe currently offer the first type only, which is why we recommend keeping an OpenAI or Azure key even when Claude is the main chat model.

Model line-up at a glance

The September 2026 line-up is easiest to read as four lanes by purpose and cost:

One cost trap on GPT-6 Astra: once a request exceeds 272K input tokens, the entire request is billed at double the input rate. The flagship's $10 / $50 is about 2.5 times GPT-5.6 Sol's promotional rate — a school wiring up the API itself will struggle to budget month to month.

Where the bill comes from

OpenAI charges developers by token, with input and output priced separately. Three realities matter for schools:

First layer: list price is not a school's spend. The $10 / $50 figure is API pricing. Edor.ai is a fixed annual subscription with AI usage included and no token charges.

Second layer: tiering beats switching vendors. Planning, tiered worksheets and draft letters to parents are well served by Terra or GPT-4.1 mini; only long reasoning needs the flagship. The platform routes by task difficulty so every click does not burn flagship rates.

Third layer: self-integration needs caps. If an IT coordinator wires up the API directly, set usage quotas and monitoring first. Without them, one runaway batch job can burn a month's budget.

Pasting a whole term's materials into the most expensive tier is not "using the context window" — it is choosing double billing. The right school approach is RAG that retrieves only the relevant passages.

How it fits inside Edor.ai

OpenAI is a native provider and the platform default. Once an administrator adds one school API key, teacher tools, the AI tutor, student conversations, knowledge-base retrieval and oral practice all work, with no per-teacher accounts. A school can switch to Azure OpenAI instead and manage data residency in its own subscription.

Two points belong in procurement papers. The key belongs to the school and can be rotated or revoked at any time. The model choice is adjustable per module — for example Chinese on Claude with everything else on OpenAI. Every call passes through four-layer safety gating and ai_audit_logs, API terms require that school data is not used for training, student personal-data fields are encrypted at rest, and conversation records are purged on the retention schedule the school sets.

For teachers

After school on a Thursday, a Primary 5 General Studies class teacher tells a colleague in the staff room: "I just generated three tiered worksheets on the platform with the everyday tier — draft in under a minute, but I still have to check the rubric column myself." Planning, question-setting, rubrics, letters to parents and meeting notes do not need the flagship.

For students

Students use AI only inside a gated, monitored environment. Tutor mode restricts the model to hints rather than answers, with the hint ceiling set by the teacher. Consumer ChatGPT has age terms, no monitoring, and no way for the school to review conversations — student use needs a gated platform.

Where to invest and where to stop

Worth it when: you need one vendor covering chat, embedding and speech together; training materials need to exist in both English and Chinese; you already hold Azure and want a compliance-friendly route; you need Cantonese speech recognition that holds up with primary pupils reading aloud and background noise.

Stop and rethink when: you treat free ChatGPT as a school solution; you paste a whole term into the flagship instead of using RAG; you wire up the API yourself without usage caps; you expect the model to know your teaching schedule and marking conventions without a knowledge base and prompt templates; you send outward-facing documents or AI marking results without teacher review — hallucination remains, especially on Hong Kong curriculum detail and regulatory wording.

Copy this into an Edor.ai teacher tool to test:

You are a Hong Kong primary General Studies teacher. Design a 30-minute inquiry activity on "Hong Kong's water resources" for Primary 5.
Output: three activity objectives mapped to learning outcomes, five guiding questions from simple to complex, a student worksheet in table form, and teacher observation points.
Constraints: write in Hong Kong English conventions; do not give away answers; if any figure is uncertain, write "teacher to verify" rather than inventing it.

The trade-off for schools

• If you need knowledge-base retrieval and Cantonese oral practice running together, keep an OpenAI or Azure OpenAI key even when your main chat model moves to another vendor.
• If you wire up the API yourself rather than using Edor.ai's fixed annual fee, set usage caps and monitoring first, and tier by task difficulty rather than sending every planning click to the flagship.
• If you are briefing a principal, say plainly that API list price is not a school's spend, and require teacher review before any outward-facing document or AI marking result goes out.