GPT-6 Astra API Now Available: $10 / $50 List Price, Codex CLI Ready | HopBase

OpenAI's next-generation flagship, GPT-6 Astra (model ID gpt-6-astra), is now live on HopBase. It is available today in the Codex Plus and Codex Pro groups — if you already have a key, just switch the model name.

How to call it

• Native on both Responses and Chat Completions: /v1/responses and /v1/chat/completions both work directly, with no changes to the OpenAI SDK.
• Codex CLI works as-is: point the model name at gpt-6-astra; no base URL or auth changes needed.
• Phased rollout: OpenAI is still releasing GPT-6 in batches, so actual availability per group is whatever GET /v1/models returns — check it before you call, rather than discovering a 404.

Pricing: at OpenAI list price

Per 1M tokens	GPT-6 Astra	GPT-5.6 Sol
Input	$10	$5
Cached input	$1	$0.50
Output	$50	$30

Cached reads cost one tenth of the input price, so agent loops and long conversations — workloads that re-read the same prefix over and over — save the most. Billing is list price times actual usage, visible in real time in the console; for enterprise volume, reach out through the pricing page.

Recommendation: prefer streaming

In our measurements, streaming calls (stream=true) return the first token in a steady 2–4 seconds. Non-streaming requests currently vary widely on the upstream side (2 seconds to 2 minutes in our tests), and tail requests can hit CDN edge timeouts.

Chat, agent and Codex workloads already stream, so they are unaffected. If your backend wraps non-streaming calls, we recommend switching to streaming and aggregating on your side for now. We are tracking upstream behaviour.

Three-minute integration

curl https://api.hop-base.com/v1/responses \
  -H "Authorization: Bearer sk-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-6-astra",
    "input": "Hello, introduce yourself",
    "stream": true
  }'

Full integration steps for Chat Completions and Codex CLI are in the model reference docs. If you already have a key, switching the model name to gpt-6-astra is all it takes.

FAQ

Can I call GPT-6 Astra right now?

Yes. It is currently open in the Codex Plus and Codex Pro groups — if you already have a key, switch the model name to gpt-6-astra. OpenAI is still rolling out in phases, so availability in other groups is whatever GET /v1/models returns; check it before you call.

What does GPT-6 Astra cost?

It is served at OpenAI list price: $10 input, $1 cached input and $50 output per million tokens. Billing is list price times actual usage, with per-request cost broken out in the console. For enterprise volume, reach out through the pricing page.

Which endpoints and tools are supported?

Both Responses (/v1/responses) and Chat Completions (/v1/chat/completions) work natively, with no changes to the OpenAI SDK. Codex CLI needs nothing but the model name changed — no base URL or auth changes.

Why do you recommend streaming?

Measured on the production path on 2026-09-05, streaming first-token latency held steady at 1.5-3.6 seconds, while non-streaming upstream response times varied widely enough that tail requests can hit CDN edge timeouts. Chat, agent and Codex workloads already stream and are unaffected; if your backend wraps non-streaming calls, switch to streaming and aggregate on your side for now.

How does it compare with GPT-5.6 and Claude Fable 5.1?

For price, latency and a failure-cost-based selection method versus GPT-5.6, see GPT-6 Astra vs GPT-5.6. Claude Fable 5.1 carries the same list price and differs on caching and protocol — see same price, different bill.

For enterprise evaluation and side-by-side model selection testing, talk to an advisor; integration steps are in the documentation.