Đăng ngày 2026-09-05 · 赵予安
Trả lời trực tiếp
openai/gpt-6-astra is listed at $10/$50 per million tokens with a 1,050,000-token context window, tiered above 272K tokens. Specs, pricing, and what is unverified. Hướng dẫn này dành cho các đội sản phẩm và nền tảng đang so sánh chất lượng model, chi phí, chính sách định tuyến, và rủi ro ra mắt.
What shipped
GPT-6 Astra is listed in the NextModel catalog as openai/gpt-6-astra: a 1,050,000-token context window, chat and Responses API support, and Thinking, Streaming, Computer Use, and Function Calling capabilities, with a maximum output of 8,192 tokens per request. OpenAI announced the model on September 3, 2026, positioning it as a frontier agent model; its announcement cites a 72.6% OSWorld computer-use score and near-perfect retrieval on long-context needle benchmarks. Call it through the same OpenAI-compatible base URL you already use for the rest of the catalog.
Pricing
Listed at $10 input and $50 output per million tokens (USD) for prompts up to 272,000 tokens, with cache hits at $1. Above 272,000 tokens the listed tier is $20/$75 with $2 cache hits. See the worked examples in our pricing explainer before sending half-million-token prompts.
Who this launch is for
The catalog tags vision, long-context, and agent use cases. Concretely: whole-repository reasoning, million-token document analysis, agent runs where the model operates a computer, and retrieval-heavy workloads where you would rather pay for context than maintain a RAG pipeline.
What is unverified
Benchmark figures quoted here are OpenAI's own published numbers, not first-party measurements. The catalog lists no latency band for this model yet, and there is no first-party CacheSafety row for gpt-6-astra in the NextModel bench table. Benchmark against your own workload before routing production traffic.
FAQ
What model id do I send for GPT-6 Astra?
Send openai/gpt-6-astra to your NextModel base URL with the OpenAI SDK you already have. Both the Chat Completions and Responses APIs are supported.
Is GPT-6 Astra billed differently above 272K tokens?
Yes. The catalog lists $10/$50 per million tokens up to 272,000 prompt tokens, and $20/$75 above that, with cache hits at $1 and $2 respectively. Price your prompt size before you paste a whole repository in.
Do I need a new integration?
No. NextModel exposes one OpenAI-compatible endpoint, so switching to Astra is a one-string model id change in your existing code.